R/metaseqr.plot.R
In pmoulos/metaseqR: An R package for the analysis and result reporting of RNA-Seq data by combining multiple statistical algorithms.
Defines functions
nat2log
log2disp
check.graphics.file
check.graphics.type
graphics.close
graphics.open
diagplot.avg.ftd
diagplot.ftd
diagplot.roc
make.venn.colorscheme
make.venn.counts
make.venn.areas
make.venn.pairs
diagplot.venn
diagplot.filtered
diagplot.de.heatmap
diagplot.volcano
diagplot.noiseq.saturation
diagplot.noiseq
diagplot.edaseq
diagplot.cor
diagplot.pairs
diagplot.mds
diagplot.boxplot
diagplot.metaseqr
Documented in
check.graphics.file
check.graphics.type
diagplot.avg.ftd
diagplot.boxplot
diagplot.cor
diagplot.de.heatmap
diagplot.edaseq
diagplot.filtered
diagplot.ftd
diagplot.mds
diagplot.metaseqr
diagplot.noiseq
diagplot.noiseq.saturation
diagplot.pairs
diagplot.roc
diagplot.venn
diagplot.volcano
graphics.close
graphics.open
log2disp
make.venn.areas
make.venn.colorscheme
make.venn.counts
make.venn.pairs
nat2log
#' Diagnostic plots for the metaseqr package
#'
#' This is the main function for producing sructured quality control and informative
#' graphs base on the results of the various steps of the metaseqR package. The
#' graphs produced span a variety of issues like good sample reproducibility
#' (Multi-Dimensional Scaling plot, biotype detection, heatmaps. diagplot.metaseqr,
#' apart from implementing certain package-specific plots, is a wrapper around
#' several diagnostic plots present in other RNA-Seq analysis packages such as
#' EDASeq and NOISeq.
#'
#' @param object a matrix or a data frame containing count data derived before or
#' after the normalization procedure, filtered or not by the metaseqR's filters
#' and/or p-value. The object can be fed to any of the \code{diagplot.metaseqr}
#' plotting systems but not every plot is meaningful. For example, it's meaningless
#' to create a \code{"biodist"} plot for a count matrix before normalization or
#' statistical testing.
#' @param sample.list the list containing condition names and the samples under
#' each condition.
#' @param annotation a data frame containing annotation elements for each row in
#' object. Usually, a subset of the annotation obtained by \code{\link{get.annotation}}
#' or a subset of possibly embedded annotation with the input counts table. This
#' parameter is optional and required only when diagplot.type is any of
#' \code{"biodetection"}, \code{"countsbio"}, \code{"saturation"}, \code{"rnacomp"},
#' \code{"readnoise"}, \code{"biodist"}, \code{"gcbias"}, \code{"lengthbias"} or
#' \code{"filtered"}.
#' @param contrast.list a named structured list of contrasts as returned by
#' \code{\link{make.contrast.list}} or just the vector of contrasts as defined in
#' the main help page of \code{\link{metaseqr}}. This parameter is optional and
#' required only when \code{diagplot.type} is any of \code{"deheatmap"},
#' \code{"volcano"} or \code{"biodist"}.
#' @param p.list a list of p-values for each contrast as obtained from any of the
#' \code{stat.*} methods of the metaseqr package. This parameter is optional and
#' required only when \code{diagplot.type} is any of \code{"deheatmap"},
#' \code{"volcano"} or \code{"biodist"}.
#' @param thresholds a list with the elements \code{"p"} and \code{"f"} which are
#' the p-value and the fold change cutoff when \code{diagplot.type="volcano"}.
#' @param diagplot.type one or more of the diagnostic plots supported in metaseqR
#' package. Many of these plots require the presence of additional package,
#' something that is checked while running the main metaseqr function. The supported
#' plots are \code{"mds"}, \code{"biodetection"}, \code{"countsbio"},
#' \code{"saturation"}, \code{"rnacomp"}, \code{"boxplot"}, \code{"gcbias"},
#' \code{"lengthbias"}, \code{"meandiff"}, \code{"meanvar"}, \code{"deheatmap"},
#' \code{"volcano"}, \code{"biodist"}, \code{"filtered"}, \code{"readnoise"},
#' \code{"venn"}, \code{"correl"}, \code{"pairwise"}. For a brief description of
#' these plots please see the main \code{\link{metaseqr}} help page.
#' @param is.norm a logical indicating whether object contains raw or normalized
#' data. It is not essential and it serves only plot annotation purposes.
#' @param output one or more R plotting device to direct the plot result to.
#' Supported mechanisms: \code{"png"}, \code{"jpg"}, \code{"bmp"}, \code{"pdf"},
#' \code{"ps"} or \code{"json"}. The latter is currently available for the creation
#' of interactive volcano plots only when reporting the output, through the
#' highcharts javascript library. The default plotting (\code{"x11"}) is not
#' supported due to instability in certain devices.
#' @param path the path to create output files.
#' @param ... further arguments to be passed to plot devices, such as parameter
#' from \code{\link{par}}.
#' @return A named list containing the file names of the produced plots. Each list
#' member is names according to the selected plotting device and is also a named
#' list, whose names are the plot types. The final contents are the file names in
#' case the plots are written to a physical location (not meaningful for \code{"x11"}).
#' @note In order to make the best out of this function, you should generally
#' provide the annotation argument as most and also the most informative plots
#' depend on this. If you don't know what is inside your counts table or how many
#' annotation elements you can provide by embedding it, it's always best to set
#' the annotation parameter of the main metaseqr function to \code{"download"} to
#' use predefined annotations that work better with the functions of the whole
#' package.
#' @author Panagiotis Moulos
#' @export
#' @examples
#' \dontrun{
#' require(DESeq)
#' data.matrix <- counts(makeExampleCountDataSet())
#' sample.list <- list(A=c("A1","A2"),B=c("B1","B2","B3"))
#' contrast <- "A_vs_B"
#' diagplot.metaseqr(data.matrix,sample.list,diagplot.type=c("mds","boxplot"))
#'
#' norm.args <- get.defaults("normalization","deseq")
#' object <- normalize.deseq(data.matrix,sample.list,norm.args)
#' diagplot.metaseqr(object,sample.list,diagplot.type="boxplot")
#'
#' p <- stat.deseq(object)
#' diagplot.metaseqr(object,sample.list,contrast.list=contrast,p.list=p,
#' diagplot.type="volcano")
#'}
diagplot.metaseqr <- function(object,sample.list,annotation=NULL,contrast.list=NULL,
p.list=NULL,thresholds=list(p=0.05,f=1),diagplot.type=c("mds","biodetection",
"countsbio","saturation","readnoise","rnacomp","correl","pairs","boxplot",
"gcbias","lengthbias","meandiff","meanvar","deheatmap","volcano","biodist",
"filtered","venn"),is.norm=FALSE,output="x11",path=NULL,...) {
# annotation should have the format internally created here... This function
# can be used outside so it must be checked at some point...
if (!is.matrix(object) && !is.data.frame(object))
stopwrap("object argument must be a matrix or data frame!")
if (is.null(annotation) && any(diagplot.type %in% c("biodetection",
"countsbio","saturation","rnacomp","readnoise","biodist","gcbias",
"lengthbias","filtered")))
stopwrap("annotation argument is needed when diagplot.type is ",
"\"biodetection\", \"countsbio\",\"saturation\",\"rnacomp\", ",
"\"readnoise\", \"biodist\", \"gcbias\", \"lengthbias\", ",
"\"filtered\" or \"venn\"!")
if (any(diagplot.type %in% c("deheatmap","volcano","biodist","venn"))) {
if (is.null(contrast.list))
stopwrap("contrast.list argument is needed when diagplot.type is ",
"\"deheatmap\",\"volcano\", \"biodist\" or \"venn\"!")
if (is.null(p.list))
stopwrap("The p argument which is a list of p-values for each ",
"contrast is needed when diagplot.type is \"deheatmap\", ",
"\"volcano\", \"biodist\" or \"venn\"!")
if (is.na(thresholds$p) || is.null(thresholds$p) || thresholds$p==1) {
warnwrap(paste("The p-value threshold when diagplot.type is ",
"\"deheatmap\", \"volcano\", \"biodist\" or \"venn\" must allow ",
"the normal plotting of DEG diagnostic plots! Setting to 0.05..."))
thresholds$p <- 0.05
}
}
if (is.null(path)) path <- getwd()
if (is.data.frame(object) && !("filtered" %in% diagplot.type))
object <- as.matrix(object)
if (any(diagplot.type %in% c("biodetection","countsbio","saturation",
"rnacomp","biodist","readnoise")))
covars <- list(
data=object,
length=annotation$end - annotation$start,
gc=annotation$gc_content,
chromosome=annotation[,1:3],
factors=data.frame(class=as.class.vector(sample.list)),
biotype=annotation$biotype,
gene_name=as.character(annotation$gene_name)
)
raw.plots <- c("mds","biodetection","countsbio","saturation","readnoise",
"correl","pairwise")
norm.plots <- c("boxplot","gcbias","lengthbias","meandiff","meanvar",
"rnacomp")
stat.plots <- c("deheatmap","volcano","biodist")
other.plots <- c("filtered")
venn.plots <- c("venn")
files <- list()
for (p in diagplot.type) {
disp(" Plotting ",p,"...")
if (p %in% raw.plots && !is.norm) {
switch(p,
mds = {
files$mds <- diagplot.mds(object,sample.list,output=output,
path=path)
},
biodetection = {
files$biodetection <- diagplot.noiseq(object,sample.list,
covars,which.plot=p,output=output,path=path,...)
},
countsbio = {
files$countsbio <- diagplot.noiseq(object,sample.list,
covars,which.plot=p,output=output,path=path,...)
},
saturation = {
fil <- diagplot.noiseq(object,sample.list,covars,
which.plot=p,output=output,path=path,...)
files$saturation$biotype <- fil[["biotype"]]
files$saturation$sample <- fil[["sample"]]
},
readnoise = {
files$readnoise <- diagplot.noiseq(object,sample.list,
covars,which.plot=p,output=output,path=path,...)
},
correl = {
files$correl$heatmap <- diagplot.cor(object,type="heatmap",
output=output,path=path,...)
files$correl$correlogram <- diagplot.cor(object,
type="correlogram",output=output,path=path,...)
},
pairwise = {
files$pairwise <- diagplot.pairs(object,output=output,
path=path)
}
)
}
if (p %in% norm.plots) {
switch(p,
boxplot = {
files$boxplot <- diagplot.boxplot(object,name=sample.list,
is.norm=is.norm,output=output,path=path,...)
},
gcbias = {
files$gcbias <- diagplot.edaseq(object,sample.list,
covar=annotation$gc_content,is.norm=is.norm,
which.plot=p,output=output,path=path,...)
},
lengthbias = {
files$lengthbias <- diagplot.edaseq(object,sample.list,
covar=annotation$end-annotation$start,is.norm=is.norm,
which.plot=p,output=output,path=path,...)
},
meandiff = {
fil <- diagplot.edaseq(object,sample.list,is.norm=is.norm,
which.plot=p,output=output,path=path,...)
for (n in names(fil)) {
if (!is.null(fil[[n]]))
files$meandiff[[n]] <- unlist(fil[[n]])
}
},
meanvar = {
fil <- diagplot.edaseq(object,sample.list,is.norm=is.norm,
which.plot=p,output=output,path=path,...)
for (n in names(fil)) {
if (!is.null(fil[[n]]))
files$meanvar[[n]] <- unlist(fil[[n]])
}
},
rnacomp = {
files$rnacomp <- diagplot.noiseq(object,sample.list,covars,
which.plot=p,output=output,is.norm=is.norm,path=path,
...)
}
)
}
if (p %in% stat.plots && is.norm) {
for (cnt in names(contrast.list)) {
disp(" Contrast: ",cnt)
samples <- names(unlist(contrast.list[[cnt]]))
mat <- as.matrix(object[,match(samples,colnames(object))])
switch(p,
deheatmap = {
files$deheatmap[[cnt]] <- diagplot.de.heatmap(mat,cnt,
output=output,path=path)
},
volcano = {
fc <- log2(make.fold.change(cnt,sample.list,object,1))
for (contrast in colnames(fc)) {
files$volcano[[contrast]] <- diagplot.volcano(
fc[,contrast],p.list[[cnt]],contrast,
fcut=thresholds$f,pcut=thresholds$p,
output=output,path=path)
}
},
biodist = {
files$biodist[[cnt]] <- diagplot.noiseq(object,
sample.list,covars,which.plot=p,output=output,
biodist.opts=list(p=p.list[[cnt]],
pcut=thresholds$p,name=cnt),path=path,...)
}
)
}
}
if (p %in% other.plots) {
switch(p,
filtered = {
files$filtered <- diagplot.filtered(object,annotation,
output=output,path=path)
}
)
}
if (p %in% venn.plots) {
switch(p,
venn = {
for (cnt in names(contrast.list)) {
disp(" Contrast: ",cnt)
if (!is.null(annotation)) {
alt.names <- as.character(annotation$gene_name)
names(alt.names) <- rownames(annotation)
}
else
alt.names <- NULL
files$venn[[cnt]] <- diagplot.venn(p.list[[cnt]],
pcut=thresholds$p,nam=cnt,output=output,path=path,
alt.names=alt.names)
}
}
)
}
}
return(files)
}
#' Boxplots wrapper for the metaseqR package
#'
#' A wrapper over the general boxplot function, suitable for matrices produced
#' and processed with the metaseqr package. Intended for internal use but can be
#' easily used as stand-alone. It can colors boxes based on group depending on
#' the name argument.
#'
#' @param mat the count data matrix.
#' @param name the names of the samples plotted on the boxplot. If \code{NULL},
#' the function check the column names of mat. If they are also \code{NULL}, sample
#' names are autogenerated. If \code{name="none"}, no sample names are plotted.
#' If name is a list, it should be the sample.list argument provided to the manin
#' metaseqr function. In that case, the boxes are colored per group.
#' @param log.it whether to log transform the values of mat or not. It can be
#' \code{TRUE}, \code{FALSE} or \code{"auto"} for auto-detection. Auto-detection
#' log transforms by default so that the boxplots are smooth and visible.
#' @param y.lim custom y-axis limits. Leave the string \code{"default"} for default
#' behavior.
#' @param is.norm a logical indicating whether object contains raw or normalized
#' data. It is not essential and it serves only plot annotation purposes.
#' @param output one or more R plotting device to direct the plot result to.
#' Supported mechanisms: \code{"x11"} (default), \code{"png"}, \code{"jpg"},
#' \code{"bmp"}, \code{"pdf"}, \code{"ps"} or \code{"json"}. The latter is
#' currently available for the creation of interactive volcano plots only when
#' reporting the output, through the highcharts javascript library (JSON for
#' boxplots not yet available).
#' @param path the path to create output files.
#' @param alt.names an optional vector of names, e.g. HUGO gene symbols, alternative
#' or complementary to the unique names of \code{f} or \code{p} (one of them must
#' be named!). It is used only in JSON output.
#' @param ... further arguments to be passed to plot devices, such as parameter
#' from \code{\link{par}}.
#' @return The filename of the boxplot produced if it's a file.
#' @export
#' @author Panagiotis Moulos
#' @examples
#' \dontrun{
#' require(DESeq)
#' data.matrix <- counts(makeExampleCountDataSet())
#' sample.list <- list(A=c("A1","A2"),B=c("B1","B2","B3"))
#' diagplot.boxplot(data.matrix,sample.list)
#'
#' norm.args <- get.defaults("normalization","deseq")
#' object <- normalize.deseq(data.matrix,sample.list,norm.args)
#' diagplot.boxplot(object,sample.list)
#'}
diagplot.boxplot <- function(mat,name=NULL,log.it="auto",y.lim="default",
is.norm=FALSE,output="x11",path=NULL,alt.names=NULL,...) {
if (is.null(path)) path <- getwd()
if (is.norm)
status<- "normalized"
else
status<- "raw"
# Need to log?
if (log.it=="auto") {
if (diff(range(mat,na.rm=TRUE))>1000)
mat <- log2disp(mat)
}
else if (log.it=="yes")
mat <- log2disp(mat)
# Define the axis limits based on user input
if (!is.numeric(y.lim) && y.lim=="default") {
min.y <- floor(min(mat))
max.y <- ceiling(max(mat))
}
else if (is.numeric(y.lim)) {
min.y <- y.lim[1]
max.y <- y.lim[2]
}
grouped <- FALSE
if (is.null(name)) {
if (is.null(colnames(mat)))
nams <- paste("Sample",1:ncol(mat),sep=" ")
else
nams <- colnames(mat)
}
else if (length(name)==1 && name=="none")
nams <- rep("",ncol(mat))
else if (is.list(name)) { # Is sample.list
nams <- unlist(name)
grouped <- TRUE
}
cols <- c("red3","green3","blue2","gold","skyblue","orange3","burlywood",
"red","blue","green","orange","darkgrey","green4","black","pink",
"brown","magenta","yellowgreen","pink4","seagreen4","darkcyan")
if (grouped) {
tmp <- as.numeric(factor(as.class.vector(name)))
b.cols <- cols[tmp]
}
else b.cols <- cols
mat.list <- list()
for (i in 1:ncol(mat))
mat.list[[i]] <- mat[,i]
names(mat.list) <- nams
if (output != "json") {
fil <- file.path(path,paste("boxplot_",status,".",output,sep=""))
graphics.open(output,fil)
if (!is.numeric(y.lim) && y.lim=="default")
b <- boxplot(mat.list,names=nams,col=b.cols,las=2,main=paste(
"Boxplot ",status,sep=""),...)
else
b <- boxplot(mat.list,names=nams,col=b.cols,ylim=c(min.y,max.y),
las=2,main=paste("Boxplot ",status,sep=""),...)
graphics.close(output)
}
else {
# Create boxplot object
b <- boxplot(mat.list,plot=FALSE)
colnames(b$stat) <- nams
# Locate the outliers
o.list <- lapply(names(mat.list),function(x,M,b) {
v <- b[,x]
o <- which(M[[x]]<v[1] | M[[x]]>v[5])
if (length(o)>0)
return(M[[x]][o])
else
return(NULL)
},mat.list,b$stat)
# Create output object
obj <- list(
x=NULL,
y=NULL,
plot=b,
samples=name,
ylims=c(min.y,max.y),
xlims=NULL,
status=status,
pcut=NULL,
fcut=NULL,
altnames=alt.names,
user=o.list
)
json <- boxplotToJSON(obj)
fil <- file.path(path,paste("boxplot_",status,".json",sep=""))
disp("Writing ",fil)
write(json,fil)
}
return(fil)
}
#' Multi-Dimensinal Scale plots or RNA-Seq samples
#'
#' Creates a Multi-Dimensional Scale plot for the given samples based on the count
#' data matrix. MDS plots are very useful for quality control as you can easily
#' see of samples of the same groups are clustered together based on the whole
#' dataset.
#'
#' @param x the count data matrix.
#' @param sample.list the list containing condition names and the samples under
#' each condition.
#' @param method which correlation method to use. Same as the method parameter in
#' \code{\link{cor}} function.
#' @param log.it whether to log transform the values of x or not.
#' @param output one or more R plotting device to direct the plot result to.
#' Supported mechanisms: \code{"x11"} (default), \code{"png"}, \code{"jpg"},
#' \code{"bmp"}, \code{"pdf"}, \code{"ps"} or \code{"json"}. The latter is
#' currently available for the creation of interactive volcano plots only when
#' reporting the output, through the highcharts javascript library.
#' @param path the path to create output files.
#' @param ... further arguments to be passed to plot devices, such as parameter
#' from \code{\link{par}}.
#' @return The filename of the MDS plot produced if it's a file.
#' @export
#' @author Panagiotis Moulos
#' @examples
#' \dontrun{
#' require(DESeq)
#' data.matrix <- counts(makeExampleCountDataSet())
#' sample.list <- list(A=c("A1","A2"),B=c("B1","B2","B3"))
#' diagplot.mds(data.matrix,sample.list)
#'}
diagplot.mds <- function(x,sample.list,method="spearman",log.it=TRUE,
output="x11",path=NULL,...) {
if (is.null(path)) path <- getwd()
classes <- as.factor(as.class.vector(sample.list))
design <- as.numeric(classes)
colspace <- c("red","blue","yellowgreen","orange","aquamarine2",
"pink2","seagreen4","brown","purple","chocolate")
pchspace <- c(20,17,15,16,8,3,2,0,1,4)
if (ncol(x)<3) {
warnwrap("MDS plot cannot be created with less than 3 samples! ",
"Skipping...")
return(NULL)
}
if (log.it)
y <- nat2log(x,base=2)
else
y <- x
d <- as.dist(0.5*(1-cor(y,method=method)))
mds.obj <- cmdscale(d,eig=TRUE,k=2)
xr <- diff(range(min(mds.obj$points[,1]),max(mds.obj$points[,1])))
yr <- diff(range(min(mds.obj$points[,2]),max(mds.obj$points[,2])))
xlim <- c(min(mds.obj$points[,1])-xr/10,max(mds.obj$points[,1])+xr/10)
ylim <- c(min(mds.obj$points[,2])-yr/10,max(mds.obj$points[,2])+yr/10)
if (output!="json") {
fil <- file.path(path,paste("mds.",output,sep=""))
if (output %in% c("pdf","ps","x11"))
graphics.open(output,fil,width=9,height=7)
else
graphics.open(output,fil,width=1024,height=768)
plot(mds.obj$points[,1],mds.obj$points[,2],
col=colspace[1:length(levels(classes))][design],
pch=pchspace[1:length(levels(classes))][design],
xlim=xlim,ylim=ylim,
main="MDS plot",xlab="MDS 1",ylab="MDS 2",
cex=0.9,cex.lab=0.9,cex.axis=0.9,cex.main=0.9)
text(mds.obj$points[,1],mds.obj$points[,2],labels=colnames(x),pos=3,
cex=0.7)
grid()
graphics.close(output)
}
else {
# Create output object
xx <- mds.obj$points[,1]
yy <- mds.obj$points[,2]
names(xx) <- names(yy) <- unlist(sample.list)
obj <- list(
x=xx,
y=yy,
plot=NULL,
samples=sample.list,
ylim=ylim,
xlim=xlim,
status=NULL,
pcut=NULL,
fcut=NULL,
altnames=NULL,
user=NULL
)
json <- mdsToJSON(obj)
fil <- file.path(path,"mds.json")
disp("Writing ",fil)
write(json,fil)
}
return(fil)
}
#' Massive X-Y, M-D correlation plots
#'
#' This function uses the read counts matrix to create pairwise correlation plots.
#' The upper diagonal of the final image contains simple scatterplots of each
#' sample against each other (log2 scale) while the lower diagonal contains
#' mean-difference plots for the same samples (log2 scale). This type of diagnostic
#' plot may not be interpretable for more than 10 samples.
#'
#' @param x the read counts matrix or data frame.
#' @param output one or more R plotting device to direct the plot result to.
#' Supported mechanisms: \code{"x11"} (default), \code{"png"}, \code{"jpg"},
#' \code{"bmp"}, \code{"pdf"} or \code{"ps"}.
#' @param path the path to create output files.
#' @param ... further arguments to be passed to plot devices, such as parameter
#' from \code{\link{par}}.
#' @return The filename of the pairwise comparisons plot produced if it's a file.
#' @export
#' @author Panagiotis Moulos
#' @examples
#' \dontrun{
#' require(DESeq)
#' data.matrix <- counts(makeExampleCountDataSet())
#' diagplot.pairs(data.matrix)
#'}
diagplot.pairs <- function(x,output="x11",path=NULL,...) {
x <- as.matrix(x)
x <- nat2log(x)
n <- ncol(x)
if (!is.null(colnames(x)))
nams <- colnames(x)
else
nams <- paste("Sample_",1:ncol(x),sep="")
if (!is.null(path))
fil <- file.path(path,paste("correlation_pairs",output,sep="."))
else
fil <- paste("correlation_pairs",output,sep=".")
if (output %in% c("pdf","ps","x11"))
graphics.open(output,fil,width=12,height=12)
else {
if (ncol(x)<=5)
graphics.open(output,fil,width=800,height=800,res=100)
else
graphics.open(output,fil,width=1024,height=1024,res=150)
}
# Setup the grid
par(mfrow=c(n,n),mar=c(1,1,1,1),oma=c(1,1,0,0),mgp=c(2,0.5,0),cex.axis=0.6,
cex.lab=0.6)
# Plot
for (i in 1:n)
{
for (j in 1:n)
{
if (i==j)
{
plot(0:10,0:10,type="n",xaxt="n",yaxt="n",xlab="",ylab="") # Diagonal
text(c(3,5,3),c(9.5,5,1),c("X-Y plots",nams[i],"M-D plots"),
cex=c(0.8,1,0.8))
arrows(6,9.5,9.5,9.5,angle=20,length=0.1,lwd=0.8,cex=0.8)
arrows(0.2,3.2,0.2,0.2,angle=20,length=0.1,lwd=0.8,cex=0.8)
}
else if (i<j) # XY plot
{
plot(x[,i],x[,j],pch=20,col="blue",cex=0.4,xlab=nams[i],
ylab=nams[j],...)
lines(lowess(x[,i],x[,j]),col="red")
cc <- paste("cor:",formatC(cor(x[,i],x[,j]),digits=3))
text(3,max(x[,j]-1),labels=cc,cex=0.7,)
#grid()
}
else if (i>j) # MD plot
{
plot((x[,i]+x[,j])/2,x[,j]-x[,i],pch=20,col="blue",cex=0.4,...)
lines(lowess((x[,i]+x[,j])/2,x[,j]-x[,i]),col="red")
#grid()
}
}
}
graphics.close(output)
return(fil)
}
#' Summarized correlation plots
#'
#' This function uses the read counts matrix to create heatmap or correlogram
#' correlation plots.
#'
#' @param mat the read counts matrix or data frame.
#' @param type create heatmap of correlogram plots.
#' @param output one or more R plotting device to direct the plot result to.
#' Supported mechanisms: \code{"x11"} (default), \code{"png"}, \code{"jpg"},
#' \code{"bmp"}, \code{"pdf"} or \code{"ps"}.
#' @param path the path to create output files.
#' @param ... further arguments to be passed to plot devices, such as parameter
#' from \code{\link{par}}.
#' @return The filename of the pairwise comparisons plot produced if it's a file.
#' @export
#' @author Panagiotis Moulos
#' @examples
#' \dontrun{
#' require(DESeq)
#' data.matrix <- counts(makeExampleCountDataSet())
#' diagplot.cor(data.matrix,type="heatmap")
#' diagplot.cor(data.matrix,type="correlogram")
#'}
diagplot.cor <- function(mat,type=c("heatmap","correlogram"),output="x11",
path=NULL,...) {
x <- as.matrix(mat)
type <- tolower(type[1])
check.text.args("type",type,c("heatmap","correlogram"))
#if (!require(corrplot) && type=="correlogram")
# stop("R package corrplot is required!")
cor.mat <- cor(mat)
if (!is.null(colnames(mat)))
colnames(cor.mat) <- colnames(mat)
if (!is.null(path))
fil <- file.path(path,paste("correlation_",type,".",output,sep=""))
else
fil <- paste("correlation_",type,".",output,sep="")
if (output %in% c("pdf","ps","x11"))
graphics.open(output,fil,width=7,height=7)
else
graphics.open(output,fil,width=640,height=640,res=100)
if (type=="correlogram")
corrplot(cor.mat,method="ellipse",order="hclust",...)
else if (type=="heatmap") {
n <- dim(cor.mat)[1]
labs <- matrix(NA,n,n)
for (i in 1:n)
for (j in 1:n)
labs[i,j] <- sprintf("%.2f",cor.mat[i,j])
if (n <= 5)
notecex <- 1.2
else if (n > 5 & n < 10)
notecex <- 0.9
else
notecex <- 0.7
heatmap.2(cor.mat,col=colorRampPalette(c("yellow","grey","blue")),
revC=TRUE,trace="none",symm=TRUE,Colv=TRUE,cellnote=labs,
keysize=1,density.info="density",notecex=notecex,cexCol=0.9,
cexRow=0.9,font.lab=2)
}
graphics.close(output)
return(fil)
}
#' Diagnostic plots based on the EDASeq package
#'
#' A wrapper around the plotting functions availale in the EDASeq normalization
#' Bioconductor package. For analytical explanation of each plot please see the
#' vignette of the EDASeq package. It is best to use this function through the
#' main plotting function \code{\link{diagplot.metaseqr}}.
#'
#' @param x the count data matrix.
#' @param sample.list the list containing condition names and the samples under
#' each condition.
#' @param covar The covariate to plot counts against. Usually \code{"gc"} or
#' \code{"length"}.
#' @param is.norm a logical indicating whether object contains raw or normalized
#' data. It is not essential and it serves only plot annotation purposes.
#' @param which.plot the EDASeq package plot to generate. It can be one or more
#' of \code{"meanvar"}, \code{"meandiff"}, \code{"gcbias"} or \code{"lengthbias"}.
#' Please refer to the documentation of the EDASeq package for details on the use
#' of these plots. The \code{which.plot="lengthbias"} case is not covered by
#' EDASeq documentation, however it is similar to the GC-bias plot when the
#' covariate is the gene length instead of the GC content.
#' @param output one or more R plotting device to direct the plot result to.
#' Supported mechanisms: \code{"x11"} (default), \code{"png"}, \code{"jpg"},
#' \code{"bmp"}, \code{"pdf"} or \code{"ps"}.
#' @param path the path to create output files.
#' @param ... further arguments to be passed to plot devices, such as parameter
#' from \code{\link{par}}.
#' @return The filenames of the plot produced in a named list with names the
#' which.plot argument. If \code{output="x11"}, no output filenames are produced.
#' @export
#' @author Panagiotis Moulos
#' @examples
#' \dontrun{
#' require(DESeq)
#' data.matrix <- counts(makeExampleCountDataSet())
#' sample.list <- list(A=c("A1","A2"),B=c("B1","B2","B3"))
#' gc <- runif(nrow(data.matrix))
#' diagplot.edaseq(data.matrix,sample.list,which.plot="meandiff")
#'}
diagplot.edaseq <- function(x,sample.list,covar=NULL,is.norm=FALSE,
which.plot=c("meanvar","meandiff","gcbias","lengthbias"),output="x11",
path=NULL,...) {
if (is.null(path)) path <- getwd()
check.text.args("which.plot",which.plot,c("meanvar","meandiff","gcbias",
"lengthbias"),multiarg=TRUE)
if (is.null(covar) && which.plot %in% c("gcbias","lengthbias"))
stopwrap("\"covar\" argument is required when \"which.plot\" is ",
"\"gcbias\ or \"lengthbias\"!")
if (is.norm)
status <- "normalized"
else
status <- "raw"
if (is.null(covar)) covar <- rep(NA,nrow(x))
s <- newSeqExpressionSet(x,phenoData=AnnotatedDataFrame(
data.frame(conditions=factor(as.class.vector(sample.list)),
row.names=colnames(x))),featureData=AnnotatedDataFrame(data.frame(
gc=covar,length=covar,row.names=rownames(x))))
switch(which.plot,
meandiff = {
fil <- vector("list",length(sample.list))
names(fil) <- names(sample.list)
for (n in names(sample.list)) {
if (length(sample.list[[n]])==1) {
warnwrap("Cannot create a mean-difference plot with one ",
"sample per condition! Skipping...")
next
}
pair.matrix <- combn(1:length(sample.list[[n]]),2)
fil[[n]] <- vector("list",ncol(pair.matrix))
for (i in 1:ncol(pair.matrix)) {
s1 <- sample.list[[n]][pair.matrix[1,i]]
s2 <- sample.list[[n]][pair.matrix[2,i]]
fil[[n]][[i]] <- file.path(path,paste(which.plot,"_",
status,"_",n,"_",s1,"_",s2,".",output,sep=""))
names(fil[[n]][i]) <- paste(s1,"vs",s2,sep="_")
graphics.open(output,fil[[n]][[i]])
MDPlot(s,y=pair.matrix[,i],main=paste("MD plot for ",n," ",
status," samples ",s1," and ",s2,sep=""),cex.main=0.9)
graphics.close(output)
}
}
},
meanvar = {
fil <- vector("list",length(sample.list))
names(fil) <- names(sample.list)
for (n in names(sample.list)) {
if (length(sample.list[[n]])==1) {
warnwrap("Cannot create a mean-variance plot with one ",
"sample per condition! Skipping...")
next
}
pair.matrix <- combn(1:length(sample.list[[n]]),2)
fil[[n]] <- vector("list",ncol(pair.matrix))
for (i in 1:ncol(pair.matrix)) {
s1 <- sample.list[[n]][pair.matrix[1,i]]
s2 <- sample.list[[n]][pair.matrix[2,i]]
fil[[n]][[i]] <- file.path(path,paste(which.plot,"_",status,
"_",n,"_",s1,"_",s2,".",output,sep=""))
names(fil[[n]][i]) <- paste(s1,"vs",s2,sep="_")
graphics.open(output,fil[[n]][[i]])
suppressWarnings(meanVarPlot(s,main=paste("MV plot for ",n,
" ",status," samples ",s1," and ",s2,sep=""),
cex.main=0.9))
graphics.close(output)
}
}
},
gcbias = {
if (!output=="json") {
fil <- file.path(path,paste(which.plot,"_",status,".",output,
sep=""))
graphics.open(output,fil)
biasPlot(s,"gc",xlim=c(0.1,0.9),log=TRUE,ylim=c(0,15),
main=paste("Expression - GC content ",status,sep=""))
grid()
graphics.close(output)
}
else {
obj <- list(
x=NULL,
y=NULL,
plot=NULL,
samples=sample.list,
ylim=NULL,
xlim=NULL,
status=status,
pcut=NULL,
fcut=NULL,
altnames=NULL,
user=list(counts=x,covar=covar,covarname="GC content")
)
json <- biasPlotToJSON(obj)
fil <- file.path(path,paste(which.plot,"_",status,".json",
sep=""))
disp("Writing ",fil)
write(json,fil)
}
},
lengthbias = {
if (output!="json") {
fil <- file.path(path,paste(which.plot,"_",status,".",output,
sep=""))
graphics.open(output,fil)
biasPlot(s,"length",log=TRUE,ylim=c(0,10),
main=paste("Expression - Gene length ",status,sep=""))
grid()
graphics.close(output)
}
else {
obj <- list(
x=NULL,
y=NULL,
plot=NULL,
samples=sample.list,
ylim=NULL,
xlim=NULL,
status=status,
pcut=NULL,
fcut=NULL,
altnames=NULL,
user=list(counts=x,covar=covar,
covarname="Gene/transcript length")
)
json <- biasPlotToJSON(obj)
fil <- file.path(path,paste(which.plot,"_",status,".json",
sep=""))
disp("Writing ",fil)
write(json,fil)
}
}
)
return(fil)
}
#' Diagnostic plots based on the NOISeq package
#'
#' A wrapper around the plotting functions availale in the NOISeq Bioconductor
#' package. For analytical explanation of each plot please see the vignette of
#' the NOISeq package. It is best to use this function through the main plotting
#' function \code{\link{diagplot.metaseqr}}.
#'
#' @param x the count data matrix.
#' @param sample.list the list containing condition names and the samples under
#' each condition.
#' @param covars a list (whose annotation elements are ideally a subset of an
#' annotation data frame produced by \code{\link{get.annotation}})
#' with the following members: data (the data matrix), length (gene length), gc
#' (the gene gc_content), chromosome (a data frame with chromosome name and
#' co-ordinates), factors (a factor with the experimental condition names
#' replicated by the number of samples in each experimental condition) and biotype
#' (each gene's biotype as depicted in Ensembl-like annotations).
#' @param which.plot the NOISeq package plot to generate. It can be one or more
#' of \code{"biodetection"}, \code{"countsbio"}, \code{"saturation"},
#' \code{"rnacomp"}, \code{"readnoise"} or \code{"biodist"}. Please refer to the
#' documentation of the EDASeq package for details on the use of these plots. The
#' \code{which.plot="saturation"} case is modified to be more informative by
#' producing two kinds of plots. See \code{\link{diagplot.noiseq.saturation}}.
#' @param biodist.opts a list with the following members: p (a vector of p-values,
#' e.g. the p-values of a contrast), pcut (a unique number depicting a p-value
#' cutoff, required for the \code{"biodist"} case), name (a name for the
#' \code{"biodist"} plot, e.g. the name of the contrast.
#' @param output one or more R plotting device to direct the plot result to.
#' Supported mechanisms: \code{"x11"} (default), \code{"png"}, \code{"jpg"},
#' \code{"bmp"}, \code{"pdf"} or \code{"ps"}.
#' @param path the path to create output files.
#' @param is.norm a logical indicating whether object contains raw or normalized
#' data. It is not essential and it serves only plot annotation purposes.
#' @param ... further arguments to be passed to plot devices, such as parameter
#' from \code{\link{par}}.
#' @return The filenames of the plots produced in a named list with names the
#' \code{which.plot} argument. If \code{output="x11"}, no output filenames are
#' produced.
#' @note Please note that in case of \code{"biodist"} plots, the behavior of the
#' function is unstable, mostly due to the very specific inputs this plotting
#' function accepts in the NOISeq package. We have tried to predict unstable
#' behavior and avoid exceptions through the use of tryCatch but it's still
#' possible that you might run onto an error.
#' @export
#' @author Panagiotis Moulos
#' @examples
#' \dontrun{
#' require(DESeq)
#' data.matrix <- counts(makeExampleCountDataSet())
#' sample.list <- list(A=c("A1","A2"),B=c("B1","B2","B3"))
#' lengths <- round(1000*runif(nrow(data.matrix)))
#' starts <- round(1000*runif(nrow(data.matrix)))
#' ends <- starts + lengths
#' covars <- list(
#' data=data.matrix,
#' length=lengths,
#' gc=runif(nrow(data.matrix)),
#' chromosome=data.frame(
#' chromosome=c(rep("chr1",nrow(data.matrix)/2),rep("chr2",nrow(data.matrix)/2)),
#' start=starts,
#' end=ends
#' ),
#' factors=data.frame(class=as.class.vector(sample.list)),
#' biotype=c(rep("protein_coding",nrow(data.matrix)/2),rep("ncRNA",
#' nrow(data.matrix)/2))
#' )
#' p <- runif(nrow(data.matrix))
#' diagplot.noiseq(data.matrix,sample.list,covars=covars,
#' biodist.opts=list(p=p,pcut=0.1,name="A_vs_B"))
#'}
diagplot.noiseq <- function(x,sample.list,covars,which.plot=c("biodetection",
"countsbio","saturation","rnacomp","readnoise","biodist"),output="x11",
biodist.opts=list(p=NULL,pcut=NULL,name=NULL),path=NULL,is.norm=FALSE,
...) {
if (is.null(path)) path <- getwd()
# covars is a list of gc-content, factors, length, biotype, chromosomes,
# factors, basically copy of the noiseq object
which.plot <- tolower(which.plot[1])
check.text.args("which.plot",which.plot,c("biodetection","countsbio",
"saturation","readnoise","rnacomp","biodist"),multiarg=FALSE)
if (missing(covars))
stopwrap("\"covars\" argument is required with NOISeq specific plots!")
else {
covars$biotype <- as.character(covars$biotype)
names(covars$length) <- names(covars$gc) <-
rownames(covars$chromosome) <- names(covars$biotype) <-
rownames(x)
}
if (which.plot=="biodist") {
if (is.null(biodist.opts$p))
stopwrap("A p-value must be provided for the \"biodist\" plot!")
if (is.null(biodist.opts$pcut) || is.na(biodist.opts$pcut))
biodist.opts$pcut=0.05
}
if (is.norm)
status<- "normalized"
else
status<- "raw"
# All of these plots are NOISeq specific so we need a local NOISeq object
if (any(is.na(unique(covars$biotype))))
covars$biotype=NULL # Otherwise, it will probably crash
local.obj <- NOISeq::readData(
data=x,
length=covars$gene.length,
gc=covars$gc.content,
chromosome=covars$chromosome,
#factors=data.frame(class=covars$factors),
factors=covars$factors,
biotype=covars$biotype
)
switch(which.plot,
biodetection = {
diagplot.data <- NOISeq::dat(local.obj,type=which.plot)
samples <- unlist(sample.list)
if (output!="json") {
fil <- character(length(samples))
names(fil) <- samples
for (i in 1:length(samples)) {
fil[samples[i]] <- file.path(path,paste(which.plot,"_",
samples[i],".",output,sep=""))
if (output %in% c("pdf","ps","x11"))
graphics.open(output,fil[samples[i]],width=9,height=7)
else
graphics.open(output,fil[samples[i]],width=1024,height=768)
explo.plot(diagplot.data,samples=i)
graphics.close(output)
}
}
else {
diagplot.data.save = NOISeq::dat2save(diagplot.data)
obj <- list(
x=NULL,
y=NULL,
plot=NULL,
samples=sample.list,
ylims=NULL,
xlims=NULL,
status=status,
pcut=NULL,
fcut=NULL,
altnames=covars$gene_name,
user=list(plotdata=diagplot.data.save,covars=covars)
)
json <- bioDetectionToJSON(obj)
fil <- character(length(samples))
names(fil) <- samples
for (i in 1:length(samples)) {
fil[samples[i]] <- file.path(path,
paste(which.plot,"_",samples[i],".json",sep=""))
disp("Writing ",fil[samples[i]])
write(json[[i]],fil[samples[i]])
}
}
},
countsbio = {
samples <- unlist(sample.list)
if (output!="json") {
diagplot.data <- NOISeq::dat(local.obj,type=which.plot,
factor=NULL)
fil <- character(length(samples))
names(fil) <- samples
for (i in 1:length(samples)) {
fil[samples[i]] <- file.path(path,paste(which.plot,"_",
samples[i],".",output,sep=""))
if (output %in% c("pdf","ps","x11"))
graphics.open(output,fil[samples[i]],width=9,height=7)
else
graphics.open(output,fil[samples[i]],width=1024,
height=768)
explo.plot(diagplot.data,samples=i,plottype="boxplot")
graphics.close(output)
}
}
else {
colnames(x) <- unlist(sample.list)
obj <- list(
x=NULL,
y=NULL,
plot=NULL,
samples=sample.list,
ylims=NULL,
xlims=NULL,
status=status,
pcut=NULL,
fcut=NULL,
altnames=covars$gene_name,
user=list(counts=nat2log(x),covars=covars)
)
# Write JSON by sample
fil <- vector("list",2)
names(fil) <- c("sample","biotype")
fil[["sample"]] <- character(length(samples))
names(fil[["sample"]]) <- samples
bts <- unique(as.character(obj$user$covars$biotype))
fil[["biotype"]] <- character(length(bts))
names(fil[["biotype"]]) <- bts
json <- countsBioToJSON(obj,by="sample")
for (i in 1:length(samples)) {
fil[["sample"]][samples[i]] <- file.path(path,
paste(which.plot,"_",samples[i],".json",sep=""))
disp("Writing ",fil[["sample"]][samples[i]])
write(json[[i]],fil[["sample"]][samples[i]])
}
json <- countsBioToJSON(obj,by="biotype")
for (i in 1:length(bts)) {
fil[["biotype"]][bts[i]] <- file.path(path,
paste(which.plot,"_",bts[i],".json",sep=""))
disp("Writing ",fil[["biotype"]][bts[i]])
write(json[[i]],fil[["biotype"]][bts[i]])
}
}
},
saturation = {
# For 10 saturation points
diagplot.data <- NOISeq::dat(local.obj,k=0,ndepth=9,type=which.plot)
d2s <- NOISeq::dat2save(diagplot.data)
if (output != "json")
fil <- diagplot.noiseq.saturation(d2s,output,covars$biotype,
path=path)
else {
samples <- unlist(sample.list)
obj <- list(
x=NULL,
y=NULL,
plot=NULL,
samples=sample.list,
ylims=NULL,
xlims=NULL,
status=status,
pcut=NULL,
fcut=NULL,
altnames=covars$gene_name,
user=list(plotdata=d2s)
)
# Write JSON by sample
fil <- vector("list",2)
names(fil) <- c("sample","biotype")
fil[["sample"]] <- character(length(samples))
names(fil[["sample"]]) <- samples
json <- bioSaturationToJSON(obj,by="sample")
for (i in 1:length(samples)) {
fil[["sample"]][samples[i]] <- file.path(path,
paste(which.plot,"_",samples[i],".json",sep=""))
disp("Writing ",fil[["sample"]][samples[i]])
write(json[[i]],fil[["sample"]][samples[i]])
}
json <- bioSaturationToJSON(obj,by="biotype")
fil[["biotype"]] <- character(length(json))
names(fil[["biotype"]]) <- names(json)
for (n in names(json)) {
fil[["biotype"]][n] <- file.path(path,
paste(which.plot,"_",n,".json",sep=""))
disp("Writing ",fil[["biotype"]][n])
write(json[[n]],fil[["biotype"]][n])
}
}
},
rnacomp = {
if (ncol(local.obj)<3) {
warnwrap("RNA composition plot cannot be created with less ",
"than 3 samples! Skipping...")
return(NULL)
}
if (ncol(local.obj)>12) {
warnwrap("RNA composition plot cannot be created with more ",
"than 12 samples! Skipping...")
return(NULL)
}
diagplot.data <- NOISeq::dat(local.obj,type="cd")
fil <- file.path(path,paste(which.plot,"_",status,".",output,
sep=""))
graphics.open(output,fil)
explo.plot(diagplot.data)
grid()
graphics.close(output)
},
readnoise = {
D <- cddat(local.obj)
if (output!="json") {
fil <- file.path(path,paste(which.plot,".",output,sep=""))
graphics.open(output,fil)
cdplot(D,main="RNA-Seq reads noise")
grid()
graphics.close(output)
}
else {
colnames(D$data2plot)[2:ncol(D$data2plot)] <-
unlist(sample.list)
obj <- list(
x=NULL,
y=NULL,
plot=NULL,
samples=sample.list,
xlim=NULL,
ylim=NULL,
status=NULL,
pcut=NULL,
fcut=NULL,
altnames=NULL,
user=D$data2plot
)
json <- readNoiseToJSON(obj)
fil <- file.path(path,paste(which.plot,".json",sep=""))
disp("Writing ",fil)
write(json,fil)
}
},
biodist = { # We have to fake a noiseq object
p <- biodist.opts$p
if (is.matrix(p)) p <- p[,1]
dummy <- new("Output",
comparison=c("Dummy.1","Dummy.2"),
factor=c("class"),
k=1,
lc=1,
method="n",
replicates="biological",
results=list(
data.frame(
Dummy.1=rep(1,length(p)),
Dummy.2=rep(1,length(p)),
M=rep(1,length(p)),
D=rep(1,length(p)),
prob=as.numeric(p),
ranking=rep(1,length(p)),
Length=rep(1,length(p)),
GC=rep(1,length(p)),
Chrom=as.character(covars$chromosome[,1]),
GeneStart=covars$chromosome[,2],
GeneEnd=covars$chromosome[,3],
Biotype=covars$biotype
)
),
nss=5,
pnr=0.2,
v=0.02
)
if (!is.null(biodist.opts$name))
fil <- file.path(path,paste(which.plot,"_",biodist.opts$name,
".",output,sep=""))
else
fil <- file.path(path,paste(which.plot,".",output,sep=""))
if (output %in% c("pdf","ps","x11"))
graphics.open(output,fil,width=10,height=6)
else
graphics.open(output,fil,width=1024,height=640)
tryCatch( # A lot of times, there is a problem with this function
DE.plot(dummy,chromosomes=NULL,q=biodist.opts$pcut,
graphic="distr"),
error=function(e) {
disp(" Known problem with NOISeq and external ",
"p-values detected! Trying to make a plot with ",
"alternative p-values (median of p-value ",
"distribution)...")
fil="error"
tryCatch(
DE.plot(dummy,chromosomes=NULL,
q=quantile(biodist.opts$p,0.5),
graphic="distr"),
error=function(e) {
disp(" Cannot create DEG biotype plot! This ",
"is not related to a problem with the ",
"results. Excluding...")
fil="error"
},
finally=""
)
},
finally=""
)
graphics.close(output)
}
)
return(fil)
}
#' Simpler implementation of saturation plots inspired from NOISeq package
#'
#' Helper function for \code{\link{diagplot.noiseq}} to plot feature detection
#' saturation as presented in the NOISeq package vignette. It has two main outputs:
#' a set of figures, one for each input sample depicting the saturation for each
#' biotype and one single multiplot which depicts the saturation of all samples
#' for each biotype. It expands the saturation plots of NOISeq by allowing more
#' samples to be examined in a simpler way. Don't use this function directly. Use
#' either \code{\link{diagplot.metaseqr}} or \code{\link{diagplot.noiseq}}.
#'
#' @param x the count data matrix.
#' @param o one or more R plotting device to direct the plot result to. Supported
#' mechanisms: \code{"x11"} (default), \code{"png"}, \code{"jpg"}, \code{"bmp"},
#' \code{"pdf"} or \code{"ps"}.
#' @param tb the vector of biotypes, one for each row of x.
#' @param path the path to create output files.
#' @return The filenames of the plots produced in a named list with names the
#' \code{which.plot} argument. If \code{output="x11"}, no output filenames are
#' produced.
#' @author Panagiotis Moulos
#' @examples
#' \dontrun{
#' require(DESeq)
#' data.matrix <- counts(makeExampleCountDataSet())
#' biotype=c(rep("protein_coding",nrow(data.matrix)/2),rep("ncRNA",
#' nrow(data.matrix)/2))
#' diagplot.noiseq.saturation(data.matrix,"x11",biotype)
#'}
diagplot.noiseq.saturation <- function(x,o,tb,path=NULL) {
if (is.null(path)) path <- getwd()
if (length(unique(tb))==1) {
warnwrap("Saturation plot cannot be created with only one biotype! ",
"Skipping...")
return(NULL)
}
total.biotypes <- table(tb)
the.biotypes <- names(tb)
biotypes <- colnames(x[[1]][,2:ncol(x[[1]])])
colspace <- c("red3","green4","blue2","orange3","burlywood",
"lightpink4","gold","skyblue","red2","green2","firebrick3",
"orange4","yellow4","skyblue3","tan4","gray40",
"brown2","darkgoldenrod","cyan3","coral2","cadetblue",
"bisque3","blueviolet","chocolate3","darkkhaki","dodgerblue")
pchspace <- c(rep(c(15,16,17,18),6),15)
# Plot all biotypes per sample
f.sample <- character(length(names(x)))
names(f.sample) <- names(x)
for (n in names(x)) {
f.sample[n] <- file.path(path,paste("saturation_",n,".",o,sep=""))
if (o %in% c("pdf","ps","x11"))
graphics.open(o,f.sample[n],width=10,height=7)
else
graphics.open(o,f.sample[n],width=1024,height=800)
y <- x[[n]]
sep <- match(c("global","protein_coding"),colnames(y))
yab <- cbind(y[,"depth"],y[,sep])
ynab <- y[,-sep]
colnames(yab)[1] <- colnames(ynab)[1] <- "depth"
xlim <- range(y[,"depth"])
ylim.ab <- range(yab[,2:ncol(yab)])
ylim.nab <- range(ynab[,2:ncol(ynab)])
par(cex.axis=0.9,cex.main=1,cex.lab=0.9,font.lab=2,font.axis=2,pty="m",
lty=2,lwd=1.5,mfrow=c(1,2))
plot.new()
plot.window(xlim,ylim.nab)
axis(1,at=pretty(xlim,10),labels=as.character(pretty(xlim,10)/1e+6))
axis(2,at=pretty(ylim.nab,10))
title(main="Non abundant biotype detection saturation",
xlab="Depth in millions of reads",ylab="Detected features")
co <- 0
for (b in biotypes) {
co <- co + 1
if (b=="global" || b=="protein_coding") {
# Silently do nothing
}
else {
lines(ynab[,"depth"],ynab[,b],col=colspace[co])
points(ynab[,"depth"],ynab[,b],pch=pchspace[co],
col=colspace[co],cex=1)
}
}
grid()
graphics::legend(
x="topleft",legend=colnames(ynab)[2:ncol(ynab)],xjust=1,yjust=0,
box.lty=0,x.intersp=0.5,cex=0.6,text.font=2,
col=colspace[1:(ncol(ynab)-1)],pch=pchspace[1:(ncol(ynab)-1)]
)
plot.new()
plot.window(xlim,ylim.ab)
axis(1,at=pretty(xlim,10),labels=as.character(pretty(xlim,10)/1e+6))
axis(2,at=pretty(ylim.ab,10))
title(main="Abundant biotype detection saturation",
xlab="Depth in millions of reads",ylab="Detected features")
co <- 0
for (b in c("global","protein_coding")) {
co <- co + 1
lines(yab[,"depth"],yab[,b],col=colspace[co])
points(yab[,"depth"],yab[,b],pch=16,col=colspace[co],cex=1.2)
}
grid()
graphics::legend(
x="topleft",legend=c("global","protein_coding"),xjust=1,yjust=0,
box.lty=0,lty=2,x.intersp=0.5,cex=0.7,text.font=2,
col=colspace[1:2],pch=pchspace[1:2]
)
mtext(n,side=3,line=-1.5,outer=TRUE,font=2,cex=1.3)
graphics.close(o)
}
# Plot all samples per biotype
g <- make.grid(length(biotypes))
f.all <- file.path(path,paste("biotype_saturation.",o,sep=""))
if (o %in% c("pdf","ps"))
graphics.open(o,f.all,width=14,height=14)
else
graphics.open(o,f.all,width=1600,height=1600,res=150)
par(cex.axis=0.8,cex.main=0.9,cex.lab=0.8,pty="m",lty=2,lwd=1.5,mfrow=g,
mar=c(3,3,1,1),oma=c(1,1,0,0),mgp=c(2,0.5,0))
for (b in biotypes) {
y <- depth <- vector("list",length(x))
names(y) <- names(depth) <- names(x)
for (n in names(x)) {
y[[n]] <- x[[n]][,b]
depth[[n]] <- x[[n]][,"depth"]
}
y <- do.call("cbind",y)
xlim <- range(do.call("c",depth))
ylim <- range(y)
plot.new()
plot.window(xlim,ylim)
axis(1,at=pretty(xlim,5),labels=as.character(pretty(xlim,5)/1e+6),
line=0.5)
axis(2,at=pretty(ylim,5),line=0.5)
title(main=b,xlab="Depth in millions of reads",
ylab="Detected features")
co <- 0
for (n in colnames(y)) {
co <- co + 1
lines(depth[[n]],y[,n],col=colspace[co])
points(depth[[n]],y[,n],pch=pchspace[co],col=colspace[co])
}
grid()
graphics::legend(
x="bottomright",legend=colnames(y),xjust=1,yjust=0,
box.lty=0,x.intersp=0.5,
col=colspace[1:length(colnames(y))],
pch=pchspace[1:length(colnames(y))]
)
}
graphics.close(o)
return(list(sample=f.sample,biotype=f.all))
}
#' (Interactive) volcano plots of differentially expressed genes
#'
#' This function plots a volcano plot or returns a JSON string which is used to
#' render aninteractive in case of HTML reporting.
#'
#' @param f the fold changes which are to be plotted on the x-axis.
#' @param p the p-values whose -log10 transformation is going to be plotted on
#' the y-axis.
#' @param con an optional string depicting a name (e.g. the contrast name) to
#' appear in the title of the volcano diagplot.
#' @param fcut a fold change cutoff so as to draw two vertical lines indicating
#' the cutoff threshold for biological significance.
#' @param pcut a p-value cutoff so as to draw a horizontal line indicating the
#' cutoff threshold for statistical significance.
#' @param alt.names an optional vector of names, e.g. HUGO gene symbols, alternative
#' or complementary to the unique names of \code{f} or \code{p} (one of them must
#' be named!). It is used only in JSON output.
#' @param output one or more R plotting device to direct the plot result to.
#' Supported mechanisms: \code{"x11"} (default), \code{"png"}, \code{"jpg"},
#' \code{"bmp"}, \code{"pdf"}, \code{"ps"} or \code{"json"}. The latter is currently
#' available for the creation of interactive volcano plots only when reporting the
#' output, through the highcharts javascript library.
#' @param path the path to create output files.
#' @param ... further arguments to be passed to plot devices, such as parameter
#' from \code{\link{par}}.
#' @return The filenames of the plots produced in a named list with names the
#' \code{which.plot} argument. If \code{output="x11"}, no output filenames are
#' produced.
#' @export
#' @author Panagiotis Moulos
#' @examples
#' \dontrun{
#' require(DESeq)
#' data.matrix <- counts(makeExampleCountDataSet())
#' sample.list <- list(A=c("A1","A2"),B=c("B1","B2","B3"))
#' contrast <- "A_vs_B"
#' M <- norm.edger(data.matrix,sample.list)
#' p <- stat.edger(M,sample.list,contrast)
#' ma <- apply(M[,sample.list$A],1,mean)
#' mb <- apply(M[,sample.list$B],1,mean)
#' f <- log2(ifelse(mb==0,1,mb)/ifelse(ma==0,1,ma))
#' diagplot.volcano(f,p,con=contrast)
#' j <- diagplot.volcano(f,p,con=contrast,output="json")
#'}
diagplot.volcano <- function(f,p,con=NULL,fcut=1,pcut=0.05,alt.names=NULL,
output="x11",path=NULL,...) { # output can be json here...
## Check rjson
#if ("json" %in% output && !require(rjson))
# stopwrap("R package rjson is required to create interactive volcano plot!")
if (is.null(path)) path <- getwd()
if (is.null(con))
con <- conn <- ""
else {
conn <- con
con <- paste("for ",con)
}
fil <- file.path(path,paste("volcano_plot_",conn,".",output,sep=""))
if (output!="json") {
if (output %in% c("pdf","ps","x11"))
graphics.open(output,fil,width=8,height=10)
else
graphics.open(output,fil,width=768,height=1024,res=100)
}
rem <- which(is.na(p))
if (length(rem)>0) {
p <- p[-rem]
f <- f[-rem]
if (!is.null(alt.names))
alt.names <- alt.names[-rem]
}
# Fix problem with extremely low p-values, only for display purposes though
p.zero <- which(p==0)
if (length(p.zero)>0)
p[p.zero] <- runif(length(p.zero),0,1e-256)
xlim <- c(-max(abs(f)),max(abs(f)))
ylim <- c(0,ceiling(-log10(min(p))))
up <- which(f>=fcut & p<pcut)
down <- which(f<=-fcut & p<pcut)
u <- union(up,down)
alt.names.neutral <- NULL
if (length(u)>0) {
ff <- f[-u]
pp <- p[-u]
if (!is.null(alt.names))
alt.names.neutral <- alt.names[-u]
}
else {
ff <- f
pp <- p
if (!is.null(alt.names))
alt.names.neutral <- alt.names
}
if (output!="json") {
par(cex.main=1.1,cex.lab=1.1,cex.axis=1.1,font.lab=2,font.axis=2,
pty="m",lwd=1.5)
plot.new()
plot.window(xlim,ylim)
axis(1,at=pretty(xlim,10),labels=as.character(pretty(xlim,10)))
axis(2,at=pretty(ylim,10))
title(paste(main="Volcano plot",con),
xlab="Fold change",ylab="-log10(p-value)")
points(ff,-log10(pp),pch=20,col="blue2",cex=0.9)
points(f[down],-log10(p[down]),pch=20,col="green3",cex=0.9)
points(f[up],-log10(p[up]),pch=20,col="red2",cex=0.9)
abline(h=-log10(pcut),lty=4)
abline(v=-fcut,lty=2)
abline(v=fcut,lty=2)
grid()
graphics::legend(
x="topleft",
legend=c("up-regulated","down-regulated","unregulated",
"p-value threshold","fold change threshold"),
col=c("red2","green3","blue1","black","black"),
pch=c(20,20,20,NA,NA),lty=c(NA,NA,NA,4,2),
xjust=1,yjust=0,box.lty=0,x.intersp=0.5,cex=0.8,text.font=2
)
graphics.close(output)
}
else {
obj <- list(
x=f,
y=p,
plot=NULL,
samples=NULL,
xlim=xlim,
ylim=ylim,
status=NULL,
pcut=pcut,
fcut=fcut,
altnames=alt.names,
user=list(up=up,down=down,unf=ff,unp=pp,ualt=alt.names.neutral,
con=con)
)
#json <- volcanoToJSON(obj)
#fil <- file.path(path,paste("volcano_",con,".json",sep=""))
#write(json,fil)
fil <- volcanoToJSON(obj)
}
return(fil)
}
#' Diagnostic heatmap of differentially expressed genes
#'
#' This function plots a heatmap of the differentially expressed genes produced
#' by the metaseqr workflow, useful for quality control, e.g. whether samples
#' belonging to the same group cluster together.
#'
#' @param x the data matrix to create a heatmap for.
#' @param con an optional string depicting a name (e.g. the contrast name) to
#' appear in the title of the volcano plot.
#' @param output one or more R plotting device to direct the plot result to.
#' Supported mechanisms: \code{"x11"} (default), \code{"png"}, \code{"jpg"},
#' \code{"bmp"}, \code{"pdf"}, \code{"ps"}.
#' @param path the path to create output files.
#' @param ... further arguments to be passed to plot devices, such as parameter
#' from \code{\link{par}}.
#' @return The filenames of the plots produced in a named list with names the
#' \code{which.plot} argument. If \code{output="x11"}, no output filenames are
#' produced.
#' @export
#' @author Panagiotis Moulos
#' @examples
#' \dontrun{
#' require(DESeq)
#' data.matrix <- counts(makeExampleCountDataSet())
#' sample.list <- list(A=c("A1","A2"),B=c("B1","B2","B3"))
#' contrast <- "A_vs_B"
#' M <- norm.edger(data.matrix,sample.list)
#' p <- stat.edger(M,sample.list,contrast)
#' diagplot.de.heatmap(data.matrix[p[[1]]<0.05])
#'}
diagplot.de.heatmap <- function(x,con=NULL,output="x11",path=NULL,...) {
if (is.null(path)) path <- getwd()
if (is.null(con))
con <- conn <- ""
else {
conn <- con
con <- paste("for ",con)
}
y <- nat2log(x,2,1)
# First plot the normal image
fil <- file.path(path,paste("de_heatmap_",conn,".",output,sep=""))
if (output %in% c("pdf","ps","x11"))
graphics.open(output,fil,width=10,height=10)
else
graphics.open(output,fil,width=800,height=800)
heatmap.2(y,trace="none",col=bluered(16),labRow="",cexCol=0.9,keysize=1,
font.lab=2,main=paste("DEG heatmap",con),cex.main=0.9)
graphics.close(output)
## Then the "interactive" using sendplot
#xy.labels <- list(normalized_counts=x,log2_normalized_counts=y)
#x.labels <- data.frame(
# label=colnames(x),
# description=paste("Sample",colnames(x))
#)
#y.labels <- data.frame(
# label=rownames(x),
# description=paste("Gene ID:",rownames(x))
#)
#suppressWarnings(heatmap.send(
# y,
# distfun=dist,
# hclustfun=hclust,
# MainColor=bluered(16),
# labRow="",
# labCol=NULL,
# keep.dendro=TRUE,
# x.labels=x.labels,
# y.labels=y.labels,
# xy.labels=xy.labels,
# image.size="2048x4096",
# fname.root=paste("iframe_de_heatmap_",conn,sep=""),
# dir=paste(path,.Platform$file.sep,sep=""),
# header="v3",
# window.size="2048x4192"
#))
return(fil)
}
#' Diagnostic plot for filtered genes
#'
#' This function plots a grid of four graphs depicting: in the first row, the
#' numbers of filtered genes per chromosome in the first column and per biotype
#' in the second column. In the second row, the percentages of filtered genes
#' per chromosome related to the whole genome in the first columns and per biotype
#' in the second column.
#'
#' @param x an annotation data frame like the ones produced by
#' \code{\link{get.annotation}}. \code{x} should be the filtered annotation
#' according to metaseqR's filters.
#' @param y an annotation data frame like the ones produced by
#' \code{\link{get.annotation}}. \code{y} should contain the total annotation
#' without the application of any metaseqr filter.
#' @param output one or more R plotting device to direct the plot result to.
#' Supported mechanisms: \code{"x11"} (default), \code{"png"}, \code{"jpg"},
#' \code{"bmp"}, \code{"pdf"} or \code{"ps"}.
#' @param path the path to create output files.
#' @param ... further arguments to be passed to plot devices, such as parameter
#' from \code{\link{par}}.
#' @return The filenames of the plots produced in a named list with names the
#' \code{which.plot} argument. If output=\code{"x11"}, no output filenames are
#' produced.
#' @export
#' @author Panagiotis Moulos
#' @examples
#' \dontrun{
#' y <- get.annotation("mm9","gene")
#' x <- y[-sample(1:nrow(y),10000),]
#' diagplot.filtered(x,y)
#'}
diagplot.filtered <- function(x,y,output="x11",path=NULL,...) {
if (output !="json") {
if (is.null(path)) path <- getwd()
fil <- file.path(path,paste("filtered_genes.",output,sep=""))
if (output %in% c("pdf","ps","x11"))
graphics.open(output,fil,width=12,height=8)
else
graphics.open(output,fil,width=1200,height=800,res=100)
chr <- table(as.character(x$chromosome))
bt <- table(as.character(x$biotype))
chr.all <- table(as.character(y$chromosome))
bt.all <- table(as.character(y$biotype))
barlab.chr <- as.character(chr)
barlab.bt <- as.character(bt)
per.chr <- chr/chr.all[names(chr)]
per.bt <- bt/bt.all[names(bt)]
# Some bug...
per.chr[per.chr>1] <- 1
per.bt[per.bt>1] <- 1
#
suppressWarnings(per.chr.lab <- paste(formatC(100*per.chr,digits=1,
format="f"),"%",sep=""))
suppressWarnings(per.bt.lab <- paste(formatC(100*per.bt,digits=1,
format="f"),"%",sep=""))
par(mfrow=c(2,2),mar=c(1,4,2,1),oma=c(1,1,1,1))
# Chromosomes
barx.chr <- barplot(chr,space=0.5,
ylim=c(0,max(chr)+ceiling(max(chr)/10)),yaxt="n",xaxt="n",
plot=FALSE)
plot.new()
plot.window(xlim=c(0,ceiling(max(barx.chr))),
ylim=c(0,max(chr)+ceiling(max(chr)/10)),mar=c(1,4,1,1))
axis(2,at=pretty(0:(max(chr)+ceiling(max(chr)/10))),cex.axis=0.9,padj=1,
font=2)
text(x=barx.chr,y=chr,label=barlab.chr,cex=0.7,font=2,col="green3",
adj=c(0.5,-1.3))
title(main="Filtered genes per chromosome",cex.main=1.1)
mtext(side=2,text="Number of genes",line=2,cex=0.9,font=2)
grid()
barplot(chr,space=0.5,ylim=c(0,max(chr)+ceiling(max(chr)/10)),
col="blue3",border="yellow3",yaxt="n",xaxt="n",font=2,add=TRUE)
# Biotypes
barx.bt <- barplot(bt,space=0.5,ylim=c(0,max(bt)+ceiling(max(bt)/10)),
yaxt="n",xaxt="n",plot=FALSE)
plot.new()
plot.window(xlim=c(0,ceiling(max(barx.bt))),
ylim=c(0,max(bt)+ceiling(max(bt)/10)),mar=c(1,4,1,1))
axis(2,at=pretty(0:(max(bt)+ceiling(max(bt)/10))),cex.axis=0.9,padj=1,
font=2)
text(x=barx.bt,y=bt,label=barlab.bt,cex=0.7,font=2,col="blue",
adj=c(0.5,-1.3),xpd=TRUE)
title(main="Filtered genes per biotype",cex.main=1.1)
mtext(side=2,text="Number of genes",line=2,cex=0.9,font=2)
grid()
barplot(bt,space=0.5,ylim=c(0,max(bt)+ceiling(max(bt)/10)),col="red3",
border="yellow3",yaxt="n",xaxt="n",font=2,add=TRUE)
# Chromosome percentage
barx.per.chr <- barplot(per.chr,space=0.5,ylim=c(0,max(per.chr)),
yaxt="n",xaxt="n",plot=FALSE)
plot.new()
par(mar=c(9,4,1,1))
plot.window(xlim=c(0,max(barx.per.chr)),ylim=c(0,max(per.chr)))
#axis(1,at=barx.per.chr,labels=names(per.chr),cex.axis=0.9,font=2,
# tcl=-0.3,col="lightgrey",las=2)
axis(1,at=barx.per.chr,labels=FALSE,tcl=-0.3,col="lightgrey")
axis(2,at=seq(0,max(per.chr),length.out=5),labels=formatC(seq(0,
max(per.chr),length.out=5),digits=2,format="f"),cex.axis=0.9,padj=1,
font=2)
text(barx.per.chr,par("usr")[3]-max(per.chr)/17,labels=names(per.chr),
srt=45,adj=c(1,1.1),xpd=TRUE,cex=0.9,font=2)
text(x=barx.per.chr,y=per.chr,label=per.chr.lab,cex=0.7,font=2,
col="green3",adj=c(0.5,-1.3),xpd=TRUE)
mtext(side=2,text="fraction of total genes",line=2,cex=0.9,font=2)
grid()
barplot(per.chr,space=0.5,ylim=c(0,max(per.chr)),col="blue3",
border="yellow3",yaxt="n",xaxt="n",font=2,add=TRUE)
# Biotype percentage
barx.per.bt <- barplot(per.bt,space=0.5,ylim=c(0,max(per.bt)),yaxt="n",
xaxt="n",plot=FALSE)
plot.new()
par(mar=c(9,4,1,1))
plot.window(xlim=c(0,max(barx.per.bt)),ylim=c(0,max(per.bt)))
#axis(1,at=barx.per.bt,labels=names(per.bt),cex.axis=0.9,font=2,
# tcl=-0.3,col="lightgrey",las=2)
axis(1,at=barx.per.bt,labels=FALSE,tcl=-0.3,col="lightgrey")
axis(2,at=seq(0,max(per.bt),length.out=5),
labels=formatC(seq(0,max(per.bt),length.out=5),digits=2,format="f"),
cex.axis=0.9,padj=1,font=2)
text(barx.per.bt,par("usr")[3]-max(per.bt)/17,
labels=gsub("prime","'",names(per.bt)),srt=45,adj=c(1,1.1),
xpd=TRUE,cex=0.9,font=2)
text(x=barx.per.bt,y=per.bt,label=per.bt.lab,cex=0.7,font=2,col="blue",
adj=c(0.5,-1.3),xpd=TRUE)
mtext(side=2,text="fraction of total genes",line=2,cex=0.9,font=2)
grid()
barplot(per.bt,space=0.5,ylim=c(0,max(per.bt)),col="red3",
border="yellow3",yaxt="n",xaxt="n",font=2,add=TRUE)
graphics.close(output)
}
else {
obj <- list(
x=NULL,
y=NULL,
plot=NULL,
samples=NULL,
xlim=NULL,
ylim=NULL,
status=NULL,
pcut=NULL,
fcut=NULL,
altnames=NULL,
user=list(filtered=x,total=y)
)
fil <- list(chromosome=NULL,biotype=NULL)
json <- filteredToJSON(obj,by="chromosome")
fil$chromosome <- file.path(path,"filtered_genes_chromosome.json")
write(json,fil$chromosome)
json <- filteredToJSON(obj,by="biotype")
fil$biotype <- file.path(path,"filtered_genes_biotype.json")
write(json,fil$biotype)
}
return(fil)
}
#' Venn diagrams when performing meta-analysis
#'
#' This function uses the R package VennDiagram and plots an up to 5-way Venn
#' diagram depicting the common and specific to each statistical algorithm genes,
#' for each contrast. Mostly for internal use because of its main argument which
#' is difficult to construct, but can be used independently if the user grasps
#' the logic.
#'
#' @param pmat a matrix with p-values corresponding to the application of each
#' statistical algorithm. The p-value matrix must have the colnames attribute and
#' the colnames should correspond to the name of the algorithm used to fill the
#' specific column (e.g. if \code{"statistics"=c("deseq","edger","nbpseq")} then
#' \code{colnames(pmat) <-} \code{c("deseq","edger","nbpseq")}.
#' @param fcmat an optional matrix with fold changes corresponding to the application
#' of each statistical algorithm. The fold change matrix must have the colnames
#' attribute and the colnames should correspond to the name of the algorithm used
#' to fill the specific column (see the parameter \code{pmat}).
#' @param pcut a p-value cutoff for statistical significance. Defaults to
#' \code{0.05}.
#' @param fcut if \code{fcmat} is supplied, an absolute fold change cutoff to be
#' applied to \code{fcmat} to determine the differentially expressed genes for
#' each algorithm.
#' @param direction if \code{fcmat} is supplied, a keyword to denote which genes
#' to draw in the Venn diagrams with respect to their direction of regulation. It
#' can be one of \code{"dereg"} for the total of regulated genes, where
#' \code{abs(fcmat[,n])>=fcut} (default), \code{"up"} for the up-regulated genes
#' where \code{fcmat[,n]>=fcut} or \code{"down"} for the up-regulated genes where
#' \code{fcmat[,n]<=-fcut}.
#' @param nam a name to be appended to the output graphics file (if \code{"output"}
#' is not \code{"x11"}).
#' @param output one or more R plotting device to direct the plot result to.
#' Supported mechanisms: \code{"x11"} (default), \code{"png"}, \code{"jpg"},
#' \code{"bmp"}, \code{"pdf"} or \code{"ps"}.
#' @param path the path to create output files.
#' @param alt.names an optional named vector of names, e.g. HUGO gene symbols,
#' alternative or complementary to the unique gene names which are the rownames
#' of \code{pmat}. The names of the vector must be the rownames of \code{pmat}.
#' @param ... further arguments to be passed to plot devices, such as parameter
#' from \code{\link{par}}.
#' @return The filenames of the plots produced in a named list with names the
#' \code{which.plot} argument. If output=\code{"x11"}, no output filenames are
#' produced. If \code{"path"} is not \code{NULL}, a file with the intersections
#' in the Venn diagrams will be produced and written in \code{"path"}.
#' @export
#' @author Panagiotis Moulos
#' @examples
#' \dontrun{
#' p1 <- 0.001*matrix(runif(300),100,3)
#' p2 <- matrix(runif(300),100,3)
#' p <- rbind(p1,p2)
#' rownames(p) <- paste("gene",1:200,sep="_")
#' colnames(p) <- paste("method",1:3,sep="_")
#' venn.contents <- diagplot.venn(p)
#'}
diagplot.venn <- function(pmat,fcmat=NULL,pcut=0.05,fcut=0.5,
direction=c("dereg","up","down"),nam=as.character(round(1000*runif(1))),
output="x11",path=NULL,alt.names=NULL,...) {
check.text.args("direction",direction,c("dereg","up","down"))
if (is.na(pcut) || is.null(pcut) || pcut==1)
warnwrap("Illegal pcut argument! Using the default (0.05)")
algs <- colnames(pmat)
if (is.null(algs))
stopwrap("The p-value matrices must have the colnames attribute ",
"(names of statistical algorithms)!")
if (!is.null(fcmat) && (is.null(colnames(fcmat)) ||
length(intersect(colnames(pmat),colnames(fcmat)))!=length(algs)))
stopwrap("The fold change matrices must have the colnames attribute ",
"(names of statistical algorithms) and must be the same as in the ",
"p-value matrices!")
nalg <- length(algs)
if(nalg>5) {
warnwrap(paste("Cannot create a Venn diagram for more than 5 result ",
"sets! ",nalg,"found, only the first 5 will be used..."))
algs <- algs[1:5]
nalg <- 5
}
lenalias <- c("two","three","four","five")
aliases <- toupper(letters[1:nalg])
names(algs) <- aliases
genes <- rownames(pmat)
pairs <- make.venn.pairs(algs)
areas <- make.venn.areas(length(algs))
counts <- make.venn.counts(length(algs))
# Initially populate the results and counts lists so they can be used to create
# the rest of the intersections
results <- vector("list",nalg+length(pairs))
names(results)[1:nalg] <- aliases
names(results)[(nalg+1):length(results)] <- names(pairs)
if (is.null(fcmat)) {
for (a in aliases) {
results[[a]] <- genes[which(pmat[,algs[a]]<pcut)]
counts[[areas[[a]]]] <- length(results[[a]])
}
}
else {
switch(direction,
dereg = {
for (a in aliases) {
results[[a]] <-
genes[which(pmat[,algs[a]]<pcut & abs(
fcmat[,algs[a]])>=fcut)]
counts[[areas[[a]]]] <- length(results[[a]])
}
},
up = {
for (a in aliases) {
results[[a]] <-
genes[which(pmat[,algs[a]]<pcut &
fcmat[,algs[a]]>=fcut)]
counts[[areas[[a]]]] <- length(results[[a]])
}
},
down = {
for (a in aliases) {
results[[a]] <-
genes[which(pmat[,algs[a]]<pcut &
fcmat[,algs[a]]<=-fcut)]
counts[[areas[[a]]]] <- length(results[[a]])
}
}
)
}
# Now, perform the intersections
for (p in names(pairs)) {
a = pairs[[p]][1]
b = pairs[[p]][2]
results[[p]] <- intersect(results[[a]],results[[b]])
counts[[areas[[p]]]] <- length(results[[p]])
}
# And now, the Venn diagrams must be constructed
color.scheme <- make.venn.colorscheme(length(algs))
fil <- file.path(path,paste("venn_plot_",nam,".",output,sep=""))
if (output %in% c("pdf","ps","x11"))
graphics.open(output,fil,width=8,height=8)
else
graphics.open(output,fil,width=800,height=800,res=100)
switch(lenalias[length(algs)-1],
two = {
v <- draw.pairwise.venn(
area1=counts$area1,
area2=counts$area2,
cross.area=counts$cross.area,
category=paste(algs," (",aliases,")",sep=""),
lty="blank",
fill=color.scheme$fill,
cex=1.5,
cat.cex=1.3,
cat.pos=c(0,0),
cat.col=color.scheme$font,
#cat.dist=0.07,
cat.fontfamily=rep("Bookman",2)
)
},
three = {
#overrideTriple <<- TRUE
v <- draw.triple.venn(
area1=counts$area1,
area2=counts$area2,
area3=counts$area3,
n12=counts$n12,
n13=counts$n13,
n23=counts$n23,
n123=counts$n123,
category=paste(algs," (",aliases,")",sep=""),
lty="blank",
fill=color.scheme$fill,
cex=1.5,
cat.cex=1.3,
#cat.pos=c(0,0,180),
cat.col=color.scheme$font,
#cat.dist=0.07,
cat.fontfamily=rep("Bookman",3)
)
},
four = {
v <- draw.quad.venn(
area1=counts$area1,
area2=counts$area2,
area3=counts$area3,
area4=counts$area4,
n12=counts$n12,
n13=counts$n13,
n14=counts$n14,
n23=counts$n23,
n24=counts$n24,
n34=counts$n34,
n123=counts$n123,
n124=counts$n124,
n134=counts$n134,
n234=counts$n234,
n1234=counts$n1234,
category=paste(algs," (",aliases,")",sep=""),
lty="blank",
fill=color.scheme$fill,
cex=1.5,
cat.cex=1.3,
c(0.1,0.1,0.05,0.05),
cat.col=color.scheme$font,
cat.fontfamily=rep("Bookman",4)
)
},
five = {
v <- draw.quintuple.venn(
area1=counts$area1,
area2=counts$area2,
area3=counts$area3,
area4=counts$area4,
area5=counts$area5,
n12=counts$n12,
n13=counts$n13,
n14=counts$n14,
n15=counts$n15,
n23=counts$n23,
n24=counts$n24,
n25=counts$n25,
n34=counts$n34,
n35=counts$n35,
n45=counts$n45,
n123=counts$n123,
n124=counts$n124,
n125=counts$n125,
n134=counts$n134,
n135=counts$n135,
n145=counts$n145,
n234=counts$n234,
n235=counts$n235,
n245=counts$n245,
n345=counts$n345,
n1234=counts$n1234,
n1235=counts$n1235,
n1245=counts$n1245,
n1345=counts$n1345,
n2345=counts$n2345,
n12345=counts$n12345,
category=paste(algs," (",aliases,")",sep=""),
lty="blank",
fill=color.scheme$fill,
cex=1.5,
cat.cex=1.3,
cat.dist=0.1,
cat.col=color.scheme$font,
cat.fontfamily=rep("Bookman",5)
)
}
)
graphics.close(output)
# Now do something with the results
if (!is.null(path)) {
results.ex <- vector("list",length(results))
names(results.ex) <- names(results)
if (!is.null(alt.names)) {
for (n in names(results))
results.ex[[n]] <- alt.names[results[[n]]]
}
else {
for (n in names(results))
results.ex[[n]] <- results[[n]]
}
max.len <- max(sapply(results.ex,length))
for (n in names(results.ex)) {
if (length(results.ex[[n]])<max.len) {
dif <- max.len - length(results.ex[[n]])
results.ex[[n]] <- c(results.ex[[n]],rep(NA,dif))
}
}
results.ex <- do.call("cbind",results.ex)
write.table(results.ex,file=file.path(path,"..","..","lists",
paste0("venn_categories_",nam,".txt")),sep="\t",
row.names=FALSE,quote=FALSE,na="")
}
return(fil)
}
#' Helper for Venn diagrams
#'
#' This function creates a list of pairwise comparisons to be performed in order
#' to create an up to 5-way Venn diagram using the R package VennDiagram. Internal
#' use mostly.
#'
#' @param algs a vector with the names of the sets (up to length 5, if larger, it
#' will be truncated with a warning).
#' @return A list with as many pairs as the comparisons to be made for the
#' construction of the Venn diagram. The pairs are encoded with the uppercase
#' letters A through E, each one corresponding to order of the input sets.
#' @export
#' @author Panagiotis Moulos
#' @examples
#' \dontrun{
#' sets <- c("apple","pear","banana")
#' pairs <- make.venn.pairs(sets)
#'}
make.venn.pairs <- function(algs)
{
lenalias <- c("two","three","four","five")
switch(lenalias[length(algs)-1],
two = {
return(list(
AB=c("A","B")
))
},
three = {
return(list(
AB=c("A","B"),
AC=c("A","C"),
BC=c("B","C"),
ABC=c("AB","C")
))
},
four = {
return(list(
AB=c("A","B"),
AC=c("A","C"),
AD=c("A","D"),
BC=c("B","C"),
BD=c("B","D"),
CD=c("C","D"),
ABC=c("AB","C"),
ABD=c("AB","D"),
ACD=c("AC","D"),
BCD=c("BC","D"),
ABCD=c("ABC","D")
))
},
five = {
return(list(
AB=c("A","B"),
AC=c("A","C"),
AD=c("A","D"),
AE=c("A","E"),
BC=c("B","C"),
BD=c("B","D"),
BE=c("B","E"),
CD=c("C","D"),
CE=c("C","E"),
DE=c("D","E"),
ABC=c("AB","C"),
ABD=c("AB","D"),
ABE=c("AB","E"),
ACD=c("AC","D"),
ACE=c("AC","E"),
ADE=c("AD","E"),
BCD=c("BC","D"),
BCE=c("BC","E"),
BDE=c("BD","E"),
CDE=c("CD","E"),
ABCD=c("ABC","D"),
ABCE=c("ABC","E"),
ABDE=c("ABD","E"),
ACDE=c("ACD","E"),
BCDE=c("BCD","E"),
ABCDE=c("ABCD","E")
))
}
)
}
#' Helper for Venn diagrams
#'
#' This function creates a list with names the arguments of the Venn diagram
#' construction functions of the R package VennDiagram and list members the
#' internal encoding (uppercase letters A to E and combinations among then) used
#' to encode the pairwise comparisons to create the intersections needed for the
#' Venn diagrams. Internal use mostly.
#'
#' @param n the number of the sets used for the Venn diagram.
#' @return A named list, see descritpion.
#' @export
#' @author Panagiotis Moulos
#' @examples
#' \dontrun{
#' sets <- c("apple","pear","banana")
#' pairs <- make.venn.pairs(sets)
#' areas <- make.venn.areas(length(sets))
#'}
make.venn.areas <- function(n)
{
lenalias <- c("two","three","four","five")
switch(lenalias[n-1],
two = {
return(list(
A="area1",
B="area2",
AB="cross.area"
))
},
three = {
return(list(
A="area1",
B="area2",
C="area3",
AB="n12",
AC="n13",
BC="n23",
ABC="n123"
))
},
four = {
return(list(
A="area1",
B="area2",
C="area3",
D="area4",
AB="n12",
AC="n13",
AD="n14",
BC="n23",
BD="n24",
CD="n34",
ABC="n123",
ABD="n124",
ACD="n134",
BCD="n234",
ABCD="n1234"
))
},
five = {
return(list(
A="area1",
B="area2",
C="area3",
D="area4",
E="area5",
AB="n12",
AC="n13",
AD="n14",
AE="n15",
BC="n23",
BD="n24",
BE="n25",
CD="n34",
CE="n35",
DE="n45",
ABC="n123",
ABD="n124",
ABE="n125",
ACD="n134",
ACE="n135",
ADE="n145",
BCD="n234",
BCE="n235",
BDE="n245",
CDE="n345",
ABCD="n1234",
ABCE="n1235",
ABDE="n1245",
ACDE="n1345",
BCDE="n2345",
ABCDE="n12345"
))
}
)
}
#' Helper for Venn diagrams
#'
#' This function creates a list with names the arguments of the Venn diagram
#' construction functions of the R package VennDiagram and list members are
#' initially \code{NULL}. They are filled by the \code{\link{diagplot.venn}}
#' function. Internal use mostly.
#'
#' @param n the number of the sets used for the Venn diagram.
#' @return A named list, see descritpion.
#' @export
#' @author Panagiotis Moulos
#' @examples
#' \dontrun{
#' sets <- c("apple","pear","banana")
#' counts <- make.venn.counts(length(sets))
#'}
make.venn.counts <- function(n)
{
lenalias <- c("two","three","four","five")
switch(lenalias[n-1],
two = {
return(list(
area1=NULL,
area2=NULL,
cross.area=NULL
))
},
three = {
return(list(
area1=NULL,
area2=NULL,
area3=NULL,
n12=NULL,
n13=NULL,
n23=NULL,
n123=NULL
))
},
four = {
return(list(
area1=NULL,
area2=NULL,
area3=NULL,
area4=NULL,
n12=NULL,
n13=NULL,
n14=NULL,
n23=NULL,
n24=NULL,
n34=NULL,
n123=NULL,
n124=NULL,
n134=NULL,
n234=NULL,
n1234=NULL
))
},
five = {
return(list(
area1=NULL,
area2=NULL,
area3=NULL,
area4=NULL,
area5=NULL,
n12=NULL,
n13=NULL,
n14=NULL,
n15=NULL,
n23=NULL,
n24=NULL,
n25=NULL,
n34=NULL,
n35=NULL,
n45=NULL,
n123=NULL,
n124=NULL,
n125=NULL,
n134=NULL,
n135=NULL,
n145=NULL,
n234=NULL,
n235=NULL,
n245=NULL,
n345=NULL,
n1234=NULL,
n1235=NULL,
n1245=NULL,
n1345=NULL,
n2345=NULL,
n12345=NULL
))
}
)
}
#' Helper for Venn diagrams
#'
#' This function returns a list of colorschemes accroding to the number of sets.
#' Internal use.
#'
#' @param n the number of the sets used for the Venn diagram.
#' @return A list with colors for fill and font.
#' @export
#' @author Panagiotis Moulos
#' @examples
#' \dontrun{
#' sets <- c("apple","pear","banana")
#' cs <- make.venn.colorscheme(length(sets))
#'}
make.venn.colorscheme <- function(n) {
lenalias <- c("two","three","four","five")
switch(lenalias[n-1],
two = {
return(list(
fill=c("blue","orange2"),
font=c("darkblue","orange4")
))
},
three = {
return(list(
fill=c("red","green","mediumpurple"),
font=c("darkred","darkgreen","mediumpurple4")
))
},
four = {
return(list(
fill=c("red","green","mediumpurple","orange2"),
font=c("darkred","darkgreen","mediumpurple4","orange4")
))
},
five = {
return(list(
fill=c("red","green","blue","mediumpurple","orange2"),
font=c("darkred","darkgreen","darkblue","mediumpurple4",
"orange4")
))
}
)
}
#' Create basic ROC curves
#'
#' This function creates basic ROC curves using a matrix of p-values (such a matrix
#' can be derived for example from the result table of \code{\link{metaseqr}} by
#' subsetting the table to get the p-values from several algorithms) given a ground
#' truth vector for differential expression and a significance level.
#'
#' @param truth the ground truth differential expression vector. It should contain
#' only zero and non-zero elements, with zero denoting non-differentially expressed
#' genes and non-zero, differentially expressed genes. Such a vector can be obtained
#' for example by using the \code{\link{make.sim.data.sd}} function, which creates
#' simulated RNA-Seq read counts based on real data.
#' @param p a p-value matrix whose rows correspond to each element in the
#' \code{truth} vector. If the matrix has a \code{colnames} attribute, a legend
#' will be added to the plot using these names, else a set of column names will
#' be auto-generated. \code{p} can also be a list or a data frame.
#' @param sig a significance level (0 < \code{sig} <=1).
#' @param x what to plot on x-axis, can be one of \code{"fpr"}, \code{"fnr"},
#' \code{"tpr"}, \code{"tnr"} for False Positive Rate, False Negative Rate, True
#' Positive Rate and True Negative Rate respectively.
#' @param y what to plot on y-axis, same as \code{x} above.
#' @param output one or more R plotting device to direct the plot result to.
#' Supported mechanisms: \code{"x11"} (default), \code{"png"}, \code{"jpg"},
#' \code{"bmp"}, \code{"pdf"} or \code{"ps"}.
#' @param path the path to create output files.
#' @param draw boolean to determine whether to plot the curves or just return the
#' calculated values (in cases where the user wants the output for later averaging
#' for example). Defaults to \code{TRUE} (make plots).
#' @param ... further arguments to be passed to plot devices, such as parameter
#' from \code{\link{par}}.
#' @return A named list with two members. The first member is a list containing
#' the ROC statistics: \code{TP} (True Postives), \code{FP} (False Positives),
#' \code{FN} (False Negatives), \code{TN} (True Negatives), \code{FPR} (False
#' Positive Rate), \code{FNR} (False Negative Rate), \code{TPR} (True Positive
#' Rate), \code{TNR} (True Negative Rate), \code{AUC} (Area Under the Curve). The
#' second is the path to the created figure graphic.
#' @export
#' @author Panagiotis Moulos
#' @examples
#' \dontrun{
#' p1 <- 0.001*matrix(runif(300),100,3)
#' p2 <- matrix(runif(300),100,3)
#' p <- rbind(p1,p2)
#' rownames(p) <- paste("gene",1:200,sep="_")
#' colnames(p) <- paste("method",1:3,sep="_")
#' truth <- c(rep(1,40),rep(-1,40),rep(0,20),rep(1,10),rep(2,10),rep(0,80))
#' names(truth) <- rownames(p)
#' roc.obj <- diagplot.roc(truth,p)
#'}
diagplot.roc <- function(truth,p,sig=0.05,x="fpr",y="tpr",output="x11",
path=NULL,draw=TRUE,...) {
check.text.args("x",x,c("fpr","fnr","tpr","tnr","scrx","sens","spec"),
multiarg=FALSE)
check.text.args("y",y,c("fpr","fnr","tpr","tnr","scry","sens","spec"),
multiarg=FALSE)
if (is.list(p))
pmat <- do.call("cbind",p)
else if (is.data.frame(p))
pmat <- as.matrix(p)
else if (is.matrix(p))
pmat <- p
if (is.null(colnames(pmat)))
colnames(pmat) <- paste("p",1:ncol(pmat),sep="_")
ax.name <- list(
tpr="True Positive Rate",
tnr="True Negative Rate",
fpr="False Positive Rate",
fnr="False Negative Rate",
scrx="Ratio of selected",
scry="Normalized TP/(FP+FN)",
sens="Sensitivity",
spec="1 - Specificity"
)
ROC <- vector("list",ncol(pmat))
names(ROC) <- colnames(pmat)
colspace.universe <- c("red","blue","green","orange","darkgrey","green4",
"black","pink","brown","magenta","yellowgreen","pink4","seagreen4",
"darkcyan")
colspace <- colspace.universe[1:ncol(pmat)]
names(colspace) <- colnames(pmat)
eps <- min(pmat[!is.na(pmat) & pmat>0])
for (n in colnames(pmat)) {
disp("Processing ",n)
gg <- which(pmat[,n]<=sig)
psample <- -log10(pmax(pmat[gg,n],eps))
#psample <- pmat[gg,n]
size <- seq(1,length(gg))
cuts <- seq(-log10(sig),max(psample),length.out=length(gg))
#cuts <- seq(min(psample),sig,length.out=length(gg))
local.truth <- truth[gg]
S <- length(size)
TP <- FP <- FN <- TN <- FPR <- FNR <- TPR <- TNR <- SENS <- SPEC <-
SCRX <- SCRY <- numeric(S)
for (i in 1:S) {
TP[i] <- length(which(psample>cuts[i] & local.truth!=0))
FP[i] <- length(which(psample>cuts[i] & local.truth==0))
FN[i] <- length(which(psample<cuts[i] & local.truth!=0))
TN[i] <- length(which(psample<cuts[i] & local.truth==0))
## Alternatives which I keep in the code
#TP[i] <- length(intersect(names(which(psample>cuts[i])),
# names(which(local.truth!=0))))
#FP[i] <- length(intersect(names(which(psample>cuts[i])),
# names(which(local.truth==0))))
#FN[i] <- length(intersect(names(which(psample<cuts[i])),
# names(which(local.truth!=0))))
#TN[i] <- length(intersect(names(which(psample<cuts[i])),
# names(which(local.truth==0))))
#bad <- which(psample<cuts[i])
#good <- which(psample>cuts[i])
#TP[i] <- length(which(local.truth[good]!=0))
#FP[i] <- length(which(local.truth[good]==0))
#TN[i] <- length(which(local.truth[bad]==0))
#FN[i] <- length(which(local.truth[bad]!=0))
SCRX[i] <- i/S
SCRY[i] <- TP[i]/(FN[i]+FP[i])
if (FP[i]+TN[i] == 0)
FPR[i] <- 0
else
FPR[i] <- FP[i]/(FP[i]+TN[i])
FNR[i] <- FN[i]/(TP[i]+FN[i])
TPR[i] <- TP[i]/(TP[i]+FN[i])
if (TN[i]+FP[i] == 0)
TNR[i] <- 0
else
TNR[i] <- TN[i]/(TN[i]+FP[i])
SENS[i] <- TPR[i]
SPEC[i] <- 1 - TNR[i]
}
#if (all(FPR==0))
# FPR[length(FPR)] <- 1
#if (all(TNR==0)) {
# TNR[1] <- 1
# SPEC[i] <- 0
#}
ROC[[n]] <- list(TP=TP,FP=FP,FN=FN,TN=TN,
FPR=FPR,FNR=FNR,TPR=TPR,TNR=TNR,SCRX=SCRX,SCRY=SCRY/max(SCRY),
SENS=SENS,SPEC=SPEC,AUC=NULL)
}
for (n in colnames(pmat)) {
disp("Calculating AUC for ",n)
auc <- 0
for (i in 2:length(ROC[[n]][[toupper(y)]])) {
auc <- auc +
0.5*(ROC[[n]][[toupper(x)]][i]-ROC[[n]][[toupper(x)]][i-1])*
(ROC[[n]][[toupper(y)]][i]+ROC[[n]][[toupper(y)]][i-1])
}
ROC[[n]]$AUC <- abs(auc)
# There are some extreme cases, with the Intersection case for the paper
# where there are no FPs or TNs for a p-value cutoff of 0.2 (which is
# imposed in order to avoid the saturation of the ROC curves). In these
# cases, performance is virtually perfect, and the actual AUC should be
# 1. For these cases, we set it to a value between 0.95 and 0.99 to
# represent a more plausible truth.
if (ROC[[n]]$AUC==0) ROC[[n]]$AUC <- sample(seq(0.95,0.99,by=0.001),1)
}
disp("")
if (draw) {
fil <- file.path(path,paste("ROC",output,sep="."))
if (output %in% c("pdf","ps","x11"))
graphics.open(output,fil,width=8,height=8)
else
graphics.open(output,fil,width=1024,height=1024,res=100)
xlim <- c(0,1)
ylim <- c(0,1)
par(cex.axis=0.9,cex.main=1,cex.lab=0.9,font.lab=2,font.axis=2,pty="m",
lwd=1.5,lty=1)
plot.new()
plot.window(xlim,ylim)
axis(1,at=pretty(xlim,10))
axis(2,at=pretty(ylim,10))
for (n in names(ROC))
lines(ROC[[n]][[toupper(x)]],ROC[[n]][[toupper(y)]],
col=colspace[n],...)
grid()
title(xlab=ax.name[[x]],ylab=ax.name[[y]])
auc.text <- as.character(sapply(ROC,function(x)
round(x$AUC,digits=3)))
graphics::legend(x="bottomright",col=colspace,lty=1,cex=0.9,
legend=paste(names(ROC)," (AUC = ",auc.text,")",sep=""))
graphics.close(output)
}
else
fil <- NULL
return(list(ROC=ROC,truth=truth,sig.level=sig,x.axis=x,y.axis=y,path=fil))
}
## Create averaged basic ROC curves
##
## This function creates averaged basic ROC curves using a list of objects returned
## from the \code{\link{diagplot.roc}} function.
##
## @param roc.obj a list containing several lists returned from the application
## of \code{\link{diagplot.roc}} function.
## @param output one or more R plotting device to direct the plot result to.
## Supported mechanisms: \code{"x11"} (default), \code{"png"}, \code{"jpg"},
## \code{"bmp"}, \code{"pdf"} or \code{"ps"}.
## @param path the path to create output files.
## @param ... further arguments to be passed to plot devices, such as parameter
## from \code{\link{par}}.
## @return A named list with two members. The first member is a list containing
## the mean and standard deviation of ROC statistics. The second is the path to
## the created figure graphic.
## @export
## @author Panagiotis Moulos
## @examples
## \dontrun{
## # Not yet available
##}
#diagplot.avg.roc <- function(roc.obj,output="x11",path=NULL,...) {
# ax.name <- list(
# tpr="True Positive Rate",
# tnr="True Negative Rate",
# fpr="False Positive Rate",
# fnr="False Negative Rate"
# )
# stats <- names(roc.obj[[1]]$ROC)
# x <- toupper(roc.obj[[1]]$x.axis)
# y <- toupper(roc.obj[[1]]$y.axis)
# avg.ROC <- vector("list",length(stats))
# avg.ROC <- lapply(avg.ROC,function(x) {
# return(list(TP=NULL,FP=NULL,FN=NULL,TN=NULL,
# FPR=NULL,FNR=NULL,TPR=NULL,TNR=NULL,AUC=NULL))
# })
# names(avg.ROC) <- stats
# colspace.universe <- c("red","blue","green","orange","darkgrey","green4",
# "black","pink","brown","yellowgreen","magenta","pink2","seagreen4",
# "darkcyan")
# colspace <- colspace.universe[1:length(stats)]
# names(colspace) <- stats
# for (s in stats) {
# disp("Retrieving ",s)
# for (r in names(avg.ROC[[s]])) {
# if (r != "AUC") {
# #avg.ROC[[s]][[r]] <- do.call("cbind",lapply(roc.obj,
# # function(x,s,r) x$ROC[[s]][[r]],s,r))
# lapply(roc.obj,function(x,s,r) print(length(x$ROC[[s]][[r]])),s,r)
# mn <- apply(avg.ROC[[s]][[r]],1,mean)
# st <- apply(avg.ROC[[s]][[r]],1,sd)
# avg.ROC[[s]][[r]] <- list(mean=mn,sd=st)
# }
# }
# }
# disp("")
# means <- do.call("cbind",lapply(avg.ROC,function(x) x$mean))
# stds <- do.call("cbind",lapply(avg.ROC,function(x) x$sd))
# fil <- file.path(path,paste("ROC",output,sep="."))
# if (output %in% c("pdf","ps","x11"))
# graphics.open(output,fil,width=8,height=8)
# else
# graphics.open(output,fil,width=1024,height=1024,res=100)
# xlim <- c(0,1)
# ylim <- c(0,1)
# par(cex.axis=0.9,cex.main=1,cex.lab=0.9,font.lab=2,font.axis=2,pty="m",
# lwd=1.5,lty=1)
# plot.new()
# plot.window(xlim,ylim)
# axis(1,at=pretty(xlim,10))
# axis(2,at=pretty(ylim,10))
# for (n in names(ROC)) {
# lines(ROC[[n]][[x]],ROC[[n]][[y]],col=colspace[n],...)
# }
# grid()
# title(xlab=ax.name[[x]],ylab=ax.name[[y]])
# legend(x="bottomright",legend=names(ROC),col=colspace,lty=1)
# graphics.close(output)
# return(list(ROC=ROC,path=fil))
#}
#' Create False (or True) Positive (or Negative) curves
#'
#' This function creates false (or true) discovery curves using a matrix of
#' p-values (such a matrix can be derived for example from the result table of
#' \code{\link{metaseqr}} by subsetting the table to get the p-values from several
#' algorithms) given a ground truth vector for differential expression.
#'
#' @param truth the ground truth differential expression vector. It should contain
#' only zero and non-zero elements, with zero denoting non-differentially expressed
#' genes and non-zero, differentially expressed genes. Such a vector can be obtained
#' for example by using the \code{\link{make.sim.data.sd}} function, which creates
#' simulated RNA-Seq read counts based on real data. The elements of \code{truth}
#' MUST be named (e.g. each gene's name).
#' @param p a p-value matrix whose rows correspond to each element in the
#' \code{truth} vector. If the matrix has a \code{colnames} attribute, a legend
#' will be added to the plot using these names, else a set of column names will
#' be auto-generated. \code{p} can also be a list or a data frame. The p-values
#' MUST be named (e.g. each gene's name).
#' @param type what to plot, can be \code{"fpc"} for False Positive Curves
#' (default), \code{"tpc"} for True Positive Curves, \code{"fnc"} for False
#' Negative Curves or \code{"tnc"} for True Negative Curves.
#' @param N create the curves based on the top (or bottom) \code{N} ranked genes
#' (default is 2000) to be used with \code{type="fpc"} or \code{type="tpc"}.
#' @param output one or more R plotting device to direct the plot result to.
#' Supported mechanisms: \code{"x11"} (default), \code{"png"}, \code{"jpg"},
#' \code{"bmp"}, \code{"pdf"} or \code{"ps"}.
#' @param path the path to create output files.
#' @param draw boolean to determine whether to plot the curves or just return the
#' calculated values (in cases where the user wants the output for later averaging
#' for example). Defaults to \code{TRUE} (make plots).
#' @param ... further arguments to be passed to plot devices, such as parameter
#' from \code{\link{par}}.
#' @return A named list with two members: the first member (\code{ftdr}) contains
#' the values used to create the plot. The second member (\code{path}) contains
#' the path to the created figure graphic.
#' @export
#' @author Panagiotis Moulos
#' @examples
#' \dontrun{
#' p1 <- 0.001*matrix(runif(300),100,3)
#' p2 <- matrix(runif(300),100,3)
#' p <- rbind(p1,p2)
#' rownames(p) <- paste("gene",1:200,sep="_")
#' colnames(p) <- paste("method",1:3,sep="_")
#' truth <- c(rep(1,40),rep(-1,40),rep(0,10),rep(1,10),rep(2,10),rep(0,80))
#' names(truth) <- rownames(p)
#' ftd.obj <- diagplot.ftd(truth,p,N=100)
#'}
diagplot.ftd <- function(truth,p,type="fpc",N=2000,output="x11",path=NULL,
draw=TRUE,...) {
check.text.args("type",type,c("fpc","tpc","fnc","tnc"),multiarg=FALSE)
if (is.list(p))
pmat <- do.call("cbind",p)
else if (is.data.frame(p))
pmat <- as.matrix(p)
else if (is.matrix(p))
pmat <- p
else if (is.numeric(p))
pmat <- as.matrix(p)
if (is.null(colnames(pmat)))
colnames(pmat) <- paste("p",1:ncol(pmat),sep="_")
y.name <- list(
tpc="Number of True Positives",
fpc="Number of False Positives",
tnc="Number of True Negatives",
fnc="Number of False Negatives"
)
ftdr.list <- vector("list",ncol(pmat))
names(ftdr.list) <- colnames(pmat)
colspace.universe <- c("red","blue","green","orange","darkgrey","green4",
"black","pink","brown","magenta","yellowgreen","pink4","seagreen4",
"darkcyan")
colspace <- colspace.universe[1:ncol(pmat)]
names(colspace) <- colnames(pmat)
switch(type,
fpc = {
for (n in colnames(pmat)) {
disp("Processing ",n)
z <- sort(pmat[,n])
for (i in 1:N) {
nn <- length(intersect(names(z[1:i]),
names(which(truth==0))))
if (nn==0)
ftdr.list[[n]][i] <- 1
else
ftdr.list[[n]][i] <- nn
}
}
},
tpc = {
for (n in colnames(pmat)) {
disp("Processing ",n)
z <- sort(pmat[,n])
for (i in 1:N)
ftdr.list[[n]][i] <- length(intersect(names(z[1:i]),
names(which(truth!=0))))
}
},
fnc = {
for (n in colnames(pmat)) {
disp("Processing ",n)
z <- sort(pmat[,n],decreasing=TRUE)
for (i in 1:N) {
nn <- length(intersect(names(z[1:i]),
names(which(truth!=0))))
if (nn==0)
ftdr.list[[n]][i] <- 1
else
ftdr.list[[n]][i] <- nn
}
}
},
tnc = {
for (n in colnames(pmat)) {
disp("Processing ",n)
z <- sort(pmat[,n],decreasing=TRUE)
for (i in 1:N)
ftdr.list[[n]][i] <- length(intersect(names(z[1:i]),
names(which(truth==0))))
}
}
)
disp("")
if (draw) {
fil <- file.path(path,paste("FTDR_",type,".",output,sep=""))
if (output %in% c("pdf","ps","x11"))
graphics.open(output,fil,width=8,height=8)
else
graphics.open(output,fil,width=1024,height=1024,res=100)
xlim <- ylim <- c(1,N)
#ylim <- c(1,length(which(truth!=0)))
par(cex.axis=0.9,cex.main=1,cex.lab=0.9,font.lab=2,font.axis=2,pty="m",
lwd=1.5,lty=1)
plot.new()
switch(type,
fpc = {
plot.window(xlim,ylim,log="y")
axis(1,at=pretty(xlim,10))
axis(2)
for (n in names(ftdr.list)) {
lines(ftdr.list[[n]],col=colspace[n],...)
}
grid()
title(main="Selected genes vs False Positives",
xlab="Number of selected genes",ylab=y.name[[type]])
graphics::legend(x="topleft",legend=names(ftdr.list),
col=colspace,lty=1)
},
tpc = {
plot.window(xlim,ylim)
axis(1,at=pretty(xlim,10))
axis(2,at=pretty(ylim,10))
for (n in names(ftdr.list)) {
lines(ftdr.list[[n]],col=colspace[n],...)
}
grid()
title(main="Selected genes vs True Positives",
xlab="Number of selected genes",ylab=y.name[[type]])
graphics::legend(x="bottomright",legend=names(ftdr.list),
col=colspace,lty=1)
},
fnc = {
plot.window(xlim,ylim,log="y")
axis(1,at=pretty(xlim,10))
axis(2)
for (n in names(ftdr.list)) {
lines(ftdr.list[[n]],col=colspace[n],...)
}
grid()
title(main="Selected genes vs False Negatives",
xlab="Number of selected genes",ylab=y.name[[type]])
graphics::legend(x="topleft",legend=names(ftdr.list),
col=colspace,lty=1)
},
tnc = {
plot.window(xlim,ylim)
axis(1,at=pretty(xlim,10))
axis(2,at=pretty(ylim,10))
for (n in names(ftdr.list)) {
lines(ftdr.list[[n]],col=colspace[n],...)
}
grid()
title(main="Selected genes vs True Negatives",
xlab="Number of selected genes",ylab=y.name[[type]])
graphics::legend(x="bottomright",legend=names(ftdr.list),
col=colspace,lty=1)
}
)
graphics.close(output)
}
else
fil <- NULL
return(list(ftdr=ftdr.list,truth=truth,type=type,N=N,path=fil))
}
#' Create average False (or True) Discovery curves
#'
#' This function creates false (or true) discovery curves using a list containing
#' several outputs from \code{\link{diagplot.ftd}}.
#'
#' @param ftdr.obj a list with outputs from \code{\link{diagplot.ftd}}.
#' @param output one or more R plotting device to direct the plot result to.
#' Supported mechanisms: \code{"x11"} (default), \code{"png"}, \code{"jpg"},
#' \code{"bmp"}, \code{"pdf"} or \code{"ps"}.
#' @param path the path to create output files.
#' @param draw boolean to determine whether to plot the curves or just return the
#' calculated values (in cases where the user wants the output for later averaging
#' for example). Defaults to \code{TRUE} (make plots).
#' @param ... further arguments to be passed to plot devices, such as parameter
#' from \code{\link{par}}.
#' @return A named list with two members: the first member (\code{avg.ftdr})
#' contains a list with the means and the standard deviations of the averaged
#' \code{ftdr.obj} and are used to create the plot. The second member (\code{path})
#' contains the path to the created figure graphic.
#' @export
#' @author Panagiotis Moulos
#' @examples
#' \dontrun{
#' p11 <- 0.001*matrix(runif(300),100,3)
#' p12 <- matrix(runif(300),100,3)
#' p21 <- 0.001*matrix(runif(300),100,3)
#' p22 <- matrix(runif(300),100,3)
#' p31 <- 0.001*matrix(runif(300),100,3)
#' p32 <- matrix(runif(300),100,3)
#' p1 <- rbind(p11,p21)
#' p2 <- rbind(p12,p22)
#' p3 <- rbind(p31,p32)
#' rownames(p1) <- rownames(p2) <- rownames(p3) <-
#' paste("gene",1:200,sep="_")
#' colnames(p1) <- colnames(p2) <- colnames(p3) <-
#' paste("method",1:3,sep="_")
#' truth <- c(rep(1,40),rep(-1,40),rep(0,20),
#' rep(1,10),rep(2,10),rep(0,80))
#' names(truth) <- rownames(p1)
#' ftd.obj.1 <- diagplot.ftd(truth,p1,N=100,draw=FALSE)
#' ftd.obj.2 <- diagplot.ftd(truth,p2,N=100,draw=FALSE)
#' ftd.obj.3 <- diagplot.ftd(truth,p3,N=100,draw=FALSE)
#' ftd.obj <- list(ftd.obj.1,ftd.obj.2,ftd.obj.3)
#' avg.ftd.obj <- diagplot.avg.ftd(ftd.obj)
#'}
diagplot.avg.ftd <- function(ftdr.obj,output="x11",path=NULL,draw=TRUE,...) {
y.name <- list(
tpc="Number of True Positives",
fpc="Number of False Positives",
tnc="Number of True Negatives",
fnc="Number of False Negatives"
)
stats <- names(ftdr.obj[[1]]$ftdr)
type <- ftdr.obj[[1]]$type
truth <- ftdr.obj[[1]]$truth
N <- ftdr.obj[[1]]$N
avg.ftdr.obj <- vector("list",length(stats))
names(avg.ftdr.obj) <- stats
colspace.universe <- c("red","blue","green","orange","darkgrey","green4",
"black","pink","brown","magenta","yellowgreen","pink4","seagreen4",
"darkcyan")
colspace <- colspace.universe[1:length(stats)]
names(colspace) <- stats
for (s in stats) {
disp("Retrieving ",s)
avg.ftdr.obj[[s]] <- do.call("cbind",lapply(ftdr.obj,
function(x) x$ftdr[[s]]))
}
disp("")
avg.ftdr.obj <- lapply(avg.ftdr.obj,function(x) {
mn <- apply(x,1,mean)
st <- apply(x,1,sd)
return(list(mean=mn,sd=st))
})
means <- do.call("cbind",lapply(avg.ftdr.obj,function(x) x$mean))
stds <- do.call("cbind",lapply(avg.ftdr.obj,function(x) x$sd))
if (draw) {
fil <- file.path(path,paste("AVG_FTDR_",type,".",output,sep=""))
if (output %in% c("pdf","ps","x11"))
graphics.open(output,fil,width=8,height=8)
else
graphics.open(output,fil,width=1024,height=1024,res=100)
xlim <- ylim <- c(1,N)
par(cex.axis=0.9,cex.main=1,cex.lab=0.9,font.lab=2,font.axis=2,pty="m",
lwd=1.5,lty=1)
plot.new()
switch(type,
fpc = {
plot.window(xlim,ylim,log="y")
axis(1,at=pretty(xlim,10))
axis(2)
for (n in colnames(means)) {
lines(means[,n],col=colspace[n],...)
}
grid()
title(main="Selected genes vs False Positives",
xlab="Number of selected genes",ylab=y.name[[type]])
graphics::legend(x="topleft",legend=colnames(means),
col=colspace,lty=1)
},
tpc = {
plot.window(xlim,ylim)
axis(1,at=pretty(xlim,10))
axis(2,at=pretty(ylim,10))
for (n in colnames(means)) {
lines(means[,n],col=colspace[n],...)
}
grid()
title(main="Selected genes vs True Positives",
xlab="Number of selected genes",ylab=y.name[[type]])
graphics::legend(x="bottomright",legend=colnames(means),
col=colspace,lty=1)
},
fnc = {
plot.window(xlim,ylim,log="y")
axis(1,at=pretty(xlim,10))
axis(2)
for (n in colnames(means)) {
lines(means[,n],col=colspace[n],...)
}
grid()
title(main="Selected genes vs False Negatives",
xlab="Number of selected genes",ylab=y.name[[type]])
graphics::legend(x="topleft",legend=colnames(means),
col=colspace,lty=1)
},
tnc = {
plot.window(xlim,ylim)
axis(1,at=pretty(xlim,10))
axis(2,at=pretty(ylim,10))
for (n in colnames(means)) {
lines(means[,n],col=colspace[n],...)
}
grid()
title(main="Selected genes vs True Negatives",
xlab="Number of selected genes",ylab=y.name[[type]])
graphics::legend(x="bottomright",legend=colnames(means),
col=colspace,lty=1)
}
)
graphics.close(output)
}
else
fil <- NULL
return(list(avg.ftdr=list(means=means,stds=stds),path=fil))
}
#' Open plotting device
#'
#' Wrapper function to open a plotting device. Internal use only.
#'
#' @param o the plotting device, see main metaseqr function
#' @param f a filename, if the plotting device requires it (e.g. \code{"pdf"})
#' @param ... further arguments to be passed to plot devices, such as parameter
#' from \code{\link{par}}.
#' @author Panagiotis Moulos
#' @examples
#' \dontrun{
#' graphics.open("pdf","test.pdf",width=12,height=12)
#'}
graphics.open <- function(o,f,...) {
if(!check.graphics.type(o))
stopwrap("Invalid graphics output type!")
if(check.graphics.file(o) && is.null(f))
stopwrap("Please specify an output file name for your plot")
switch(o,
x11 = { dev.new(...) },
pdf = { pdf(file=f,pointsize=10,...) },
ps = { postscript(file=f,pointsize=10,...) },
png = { png(filename=f,pointsize=12,...) },
jpg = { jpeg(filename=f,pointsize=12,quality=100,...) },
bmp = { bmp(filename=f,pointsize=12,...) },
tiff = { tiff(filename=f,pointsize=12,...) }
)
}
#' Close plotting device
#'
#' Wrapper function to close a plotting device. Internal use only.
#'
#' @param o the plotting device, see main metaseqr function
#' @author Panagiotis Moulos
#' @examples
#' \dontrun{
#' graphics.close("pdf")
#'}
graphics.close <- function(o) {
if (!is.element(o,c("x11","png","jpg","tiff","bmp","pdf","ps")))
return(FALSE)
if (o!="x11")
dev.off()
}
#' Check plotting device
#'
#' Plotting device checker. Internal use only.
#'
#' @param o the plotting device, see main metaseqr function
#' @author Panagiotis Moulos
check.graphics.type <- function(o) {
if (!is.element(o,c("x11","png","jpg","tiff","bmp","pdf","ps")))
return(FALSE)
else
return(TRUE)
}
#' Check graphics file
#'
#' Graphics file checker. Internal use only.
#'
#' @param o the plotting device, see main metaseqr function
#' @author Panagiotis Moulos
check.graphics.file <- function(o) {
if (is.element(o,c("png","jpg","tiff","bmp","pdf","ps")))
return(TRUE)
else
return(FALSE)
}
#' Display value transformation
#'
#' Logarithmic transformation for display purposes. Internal use only.
#'
#' @param mat input data matrix
#' @param base logarithmic base, 2 or 10
#' @author Panagiotis Moulos
log2disp <- function(mat,base=2) {
mat[mat==0] <- 1
if (base==10)
return(log10(mat))
else
return(log2(mat))
}
#' General value transformation
#'
#' Logarithmic transformation. Internal use only.
#'
#' @param x input data matrix
#' @param base logarithmic base, 2 or 10
#' @param off offset to avoid Infinity
#' @author Panagiotis Moulos
nat2log <- function(x,base=2,off=1) {
#x[x==0] <- off
x <- x + off
if (base==2)
return(log2(x))
else
return(log10(x))
}
#' Old functions from NOISeq
#'
#' Old functions from NOISeq to create the \code{"readnoise"} plots. Internal use
#' only.
#'
#' @param input input to cddat.
#' @return a list with data to plot.
#' @note Adopted from an older version of NOISeq package (author: Sonia Tarazona).
#' @author Panagiotis Moulos
cddat <- function (input) {
if (inherits(input,"eSet") == FALSE)
stopwrap("The input data must be an eSet object.\n")
if (!is.null(assayData(input)$exprs)) {
if (ncol(assayData(input)$exprs) < 2)
stopwrap("The input object should have at least two samples.\n")
datos <- assayData(input)$exprs
}
else {
if (ncol(assayData(input)$counts) < 2)
stopwrap("The input object should have at least two samples.\n")
datos <- assayData(input)$counts
}
datos <- datos[which(rowSums(datos) > 0),]
nu <- nrow(datos) # number of detected features
qq <- 1:nu
data2plot = data.frame("%features" = 100*qq/nu)
for (i in 1:ncol(datos)) {
acumu <- 100*cumsum(sort(datos[,i],decreasing=TRUE))/sum(datos[,i])
data2plot = cbind(data2plot, acumu)
}
colnames(data2plot)[-1] = colnames(datos)
# Diagnostic test
KSpval = mostres = NULL
for (i in 1:(ncol(datos)-1)) {
for (j in (i+1):ncol(datos)) {
mostres = c(mostres, paste(colnames(datos)[c(i,j)], collapse="_"))
KSpval = c(KSpval, suppressWarnings(ks.test(datos[,i], datos[,j],
alternative = "two.sided"))$"p.value")
}
}
KSpval = p.adjust(KSpval, method = "fdr")
return(list(
"data2plot"=data2plot,
"DiagnosticTest"=data.frame("ComparedSamples"=mostres,"KSpvalue"=KSpval)
))
}
#' Old functions from NOISeq
#'
#' Old functions from NOISeq to create the \code{"readnoise"} plots. Internal use
#' only.
#' @param dat the returned list from \code{\link{cddat}}.
#' @param samples the samples to plot.
#' @param ... further arguments passed to e.g. \code{\link{par}}.
#' @return Nothing, it created the old RNA composition plot.
#' @note Adopted from an older version of NOISeq package (author: Sonia Tarazona)
#' @author Panagiotis Moulos
cdplot <- function (dat,samples=NULL,...) {
dat = dat$data2plot
if (is.null(samples)) samples <- 1:(ncol(dat)-1)
if (is.numeric(samples)) samples = colnames(dat)[samples+1]
colspace <- c("red","blue","yellowgreen","orange","aquamarine2","pink2",
"seagreen4","brown","purple","chocolate","gray10","gray30","darkblue",
"darkgreen","firebrick2","darkorange4","darkorchid","darkcyan","gold4",
"deeppink3")
if (length(samples)>length(colspace))
miscolores <- sample(colspace,length(samples),replace=TRUE)
else
miscolores <- sample(colspace,length(samples))
plot(dat[,1],dat[,samples[1]],xlab="% features",ylab="% reads",type="l",
col=miscolores[1],...)
for (i in 2:length(samples))
lines(dat[,1],dat[,samples[i]],col=miscolores[i])
graphics::legend("bottom",legend=samples,
text.col=miscolores[1:length(samples)],bty="n",lty=1,lwd=2,
col=miscolores[1:length(samples)])
}
pmoulos/metaseqR documentation built on Dec. 21, 2020, 6:19 a.m.
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
Defines functions nat2log log2disp check.graphics.file check.graphics.type graphics.close graphics.open diagplot.avg.ftd diagplot.ftd diagplot.roc make.venn.colorscheme make.venn.counts make.venn.areas make.venn.pairs diagplot.venn diagplot.filtered diagplot.de.heatmap diagplot.volcano diagplot.noiseq.saturation diagplot.noiseq diagplot.edaseq diagplot.cor diagplot.pairs diagplot.mds diagplot.boxplot diagplot.metaseqr
Documented in check.graphics.file check.graphics.type diagplot.avg.ftd diagplot.boxplot diagplot.cor diagplot.de.heatmap diagplot.edaseq diagplot.filtered diagplot.ftd diagplot.mds diagplot.metaseqr diagplot.noiseq diagplot.noiseq.saturation diagplot.pairs diagplot.roc diagplot.venn diagplot.volcano graphics.close graphics.open log2disp make.venn.areas make.venn.colorscheme make.venn.counts make.venn.pairs nat2log
#' Diagnostic plots for the metaseqr package
#'
#' This is the main function for producing sructured quality control and informative
#' graphs base on the results of the various steps of the metaseqR package. The
#' graphs produced span a variety of issues like good sample reproducibility
#' (Multi-Dimensional Scaling plot, biotype detection, heatmaps. diagplot.metaseqr,
#' apart from implementing certain package-specific plots, is a wrapper around
#' several diagnostic plots present in other RNA-Seq analysis packages such as
#' EDASeq and NOISeq.
#'
#' @param object a matrix or a data frame containing count data derived before or
#' after the normalization procedure, filtered or not by the metaseqR's filters
#' and/or p-value. The object can be fed to any of the \code{diagplot.metaseqr}
#' plotting systems but not every plot is meaningful. For example, it's meaningless
#' to create a \code{"biodist"} plot for a count matrix before normalization or
#' statistical testing.
#' @param sample.list the list containing condition names and the samples under
#' each condition.
#' @param annotation a data frame containing annotation elements for each row in
#' object. Usually, a subset of the annotation obtained by \code{\link{get.annotation}}
#' or a subset of possibly embedded annotation with the input counts table. This
#' parameter is optional and required only when diagplot.type is any of
#' \code{"biodetection"}, \code{"countsbio"}, \code{"saturation"}, \code{"rnacomp"},
#' \code{"readnoise"}, \code{"biodist"}, \code{"gcbias"}, \code{"lengthbias"} or
#' \code{"filtered"}.
#' @param contrast.list a named structured list of contrasts as returned by
#' \code{\link{make.contrast.list}} or just the vector of contrasts as defined in
#' the main help page of \code{\link{metaseqr}}. This parameter is optional and
#' required only when \code{diagplot.type} is any of \code{"deheatmap"},
#' \code{"volcano"} or \code{"biodist"}.
#' @param p.list a list of p-values for each contrast as obtained from any of the
#' \code{stat.*} methods of the metaseqr package. This parameter is optional and
#' required only when \code{diagplot.type} is any of \code{"deheatmap"},
#' \code{"volcano"} or \code{"biodist"}.
#' @param thresholds a list with the elements \code{"p"} and \code{"f"} which are
#' the p-value and the fold change cutoff when \code{diagplot.type="volcano"}.
#' @param diagplot.type one or more of the diagnostic plots supported in metaseqR
#' package. Many of these plots require the presence of additional package,
#' something that is checked while running the main metaseqr function. The supported
#' plots are \code{"mds"}, \code{"biodetection"}, \code{"countsbio"},
#' \code{"saturation"}, \code{"rnacomp"}, \code{"boxplot"}, \code{"gcbias"},
#' \code{"lengthbias"}, \code{"meandiff"}, \code{"meanvar"}, \code{"deheatmap"},
#' \code{"volcano"}, \code{"biodist"}, \code{"filtered"}, \code{"readnoise"},
#' \code{"venn"}, \code{"correl"}, \code{"pairwise"}. For a brief description of
#' these plots please see the main \code{\link{metaseqr}} help page.
#' @param is.norm a logical indicating whether object contains raw or normalized
#' data. It is not essential and it serves only plot annotation purposes.
#' @param output one or more R plotting device to direct the plot result to.
#' Supported mechanisms: \code{"png"}, \code{"jpg"}, \code{"bmp"}, \code{"pdf"},
#' \code{"ps"} or \code{"json"}. The latter is currently available for the creation
#' of interactive volcano plots only when reporting the output, through the
#' highcharts javascript library. The default plotting (\code{"x11"}) is not
#' supported due to instability in certain devices.
#' @param path the path to create output files.
#' @param ... further arguments to be passed to plot devices, such as parameter
#' from \code{\link{par}}.
#' @return A named list containing the file names of the produced plots. Each list
#' member is names according to the selected plotting device and is also a named
#' list, whose names are the plot types. The final contents are the file names in
#' case the plots are written to a physical location (not meaningful for \code{"x11"}).
#' @note In order to make the best out of this function, you should generally
#' provide the annotation argument as most and also the most informative plots
#' depend on this. If you don't know what is inside your counts table or how many
#' annotation elements you can provide by embedding it, it's always best to set
#' the annotation parameter of the main metaseqr function to \code{"download"} to
#' use predefined annotations that work better with the functions of the whole
#' package.
#' @author Panagiotis Moulos
#' @export
#' @examples
#' \dontrun{
#' require(DESeq)
#' data.matrix <- counts(makeExampleCountDataSet())
#' sample.list <- list(A=c("A1","A2"),B=c("B1","B2","B3"))
#' contrast <- "A_vs_B"
#' diagplot.metaseqr(data.matrix,sample.list,diagplot.type=c("mds","boxplot"))
#'
#' norm.args <- get.defaults("normalization","deseq")
#' object <- normalize.deseq(data.matrix,sample.list,norm.args)
#' diagplot.metaseqr(object,sample.list,diagplot.type="boxplot")
#'
#' p <- stat.deseq(object)
#' diagplot.metaseqr(object,sample.list,contrast.list=contrast,p.list=p,
#' diagplot.type="volcano")
#'}
diagplot.metaseqr <- function(object,sample.list,annotation=NULL,contrast.list=NULL,
p.list=NULL,thresholds=list(p=0.05,f=1),diagplot.type=c("mds","biodetection",
"countsbio","saturation","readnoise","rnacomp","correl","pairs","boxplot",
"gcbias","lengthbias","meandiff","meanvar","deheatmap","volcano","biodist",
"filtered","venn"),is.norm=FALSE,output="x11",path=NULL,...) {
# annotation should have the format internally created here... This function
# can be used outside so it must be checked at some point...
if (!is.matrix(object) && !is.data.frame(object))
stopwrap("object argument must be a matrix or data frame!")
if (is.null(annotation) && any(diagplot.type %in% c("biodetection",
"countsbio","saturation","rnacomp","readnoise","biodist","gcbias",
"lengthbias","filtered")))
stopwrap("annotation argument is needed when diagplot.type is ",
"\"biodetection\", \"countsbio\",\"saturation\",\"rnacomp\", ",
"\"readnoise\", \"biodist\", \"gcbias\", \"lengthbias\", ",
"\"filtered\" or \"venn\"!")
if (any(diagplot.type %in% c("deheatmap","volcano","biodist","venn"))) {
if (is.null(contrast.list))
stopwrap("contrast.list argument is needed when diagplot.type is ",
"\"deheatmap\",\"volcano\", \"biodist\" or \"venn\"!")
if (is.null(p.list))
stopwrap("The p argument which is a list of p-values for each ",
"contrast is needed when diagplot.type is \"deheatmap\", ",
"\"volcano\", \"biodist\" or \"venn\"!")
if (is.na(thresholds$p) || is.null(thresholds$p) || thresholds$p==1) {
warnwrap(paste("The p-value threshold when diagplot.type is ",
"\"deheatmap\", \"volcano\", \"biodist\" or \"venn\" must allow ",
"the normal plotting of DEG diagnostic plots! Setting to 0.05..."))
thresholds$p <- 0.05
}
}
if (is.null(path)) path <- getwd()
if (is.data.frame(object) && !("filtered" %in% diagplot.type))
object <- as.matrix(object)
if (any(diagplot.type %in% c("biodetection","countsbio","saturation",
"rnacomp","biodist","readnoise")))
covars <- list(
data=object,
length=annotation$end - annotation$start,
gc=annotation$gc_content,
chromosome=annotation[,1:3],
factors=data.frame(class=as.class.vector(sample.list)),
biotype=annotation$biotype,
gene_name=as.character(annotation$gene_name)
)
raw.plots <- c("mds","biodetection","countsbio","saturation","readnoise",
"correl","pairwise")
norm.plots <- c("boxplot","gcbias","lengthbias","meandiff","meanvar",
"rnacomp")
stat.plots <- c("deheatmap","volcano","biodist")
other.plots <- c("filtered")
venn.plots <- c("venn")
files <- list()
for (p in diagplot.type) {
disp(" Plotting ",p,"...")
if (p %in% raw.plots && !is.norm) {
switch(p,
mds = {
files$mds <- diagplot.mds(object,sample.list,output=output,
path=path)
},
biodetection = {
files$biodetection <- diagplot.noiseq(object,sample.list,
covars,which.plot=p,output=output,path=path,...)
},
countsbio = {
files$countsbio <- diagplot.noiseq(object,sample.list,
covars,which.plot=p,output=output,path=path,...)
},
saturation = {
fil <- diagplot.noiseq(object,sample.list,covars,
which.plot=p,output=output,path=path,...)
files$saturation$biotype <- fil[["biotype"]]
files$saturation$sample <- fil[["sample"]]
},
readnoise = {
files$readnoise <- diagplot.noiseq(object,sample.list,
covars,which.plot=p,output=output,path=path,...)
},
correl = {
files$correl$heatmap <- diagplot.cor(object,type="heatmap",
output=output,path=path,...)
files$correl$correlogram <- diagplot.cor(object,
type="correlogram",output=output,path=path,...)
},
pairwise = {
files$pairwise <- diagplot.pairs(object,output=output,
path=path)
}
)
}
if (p %in% norm.plots) {
switch(p,
boxplot = {
files$boxplot <- diagplot.boxplot(object,name=sample.list,
is.norm=is.norm,output=output,path=path,...)
},
gcbias = {
files$gcbias <- diagplot.edaseq(object,sample.list,
covar=annotation$gc_content,is.norm=is.norm,
which.plot=p,output=output,path=path,...)
},
lengthbias = {
files$lengthbias <- diagplot.edaseq(object,sample.list,
covar=annotation$end-annotation$start,is.norm=is.norm,
which.plot=p,output=output,path=path,...)
},
meandiff = {
fil <- diagplot.edaseq(object,sample.list,is.norm=is.norm,
which.plot=p,output=output,path=path,...)
for (n in names(fil)) {
if (!is.null(fil[[n]]))
files$meandiff[[n]] <- unlist(fil[[n]])
}
},
meanvar = {
fil <- diagplot.edaseq(object,sample.list,is.norm=is.norm,
which.plot=p,output=output,path=path,...)
for (n in names(fil)) {
if (!is.null(fil[[n]]))
files$meanvar[[n]] <- unlist(fil[[n]])
}
},
rnacomp = {
files$rnacomp <- diagplot.noiseq(object,sample.list,covars,
which.plot=p,output=output,is.norm=is.norm,path=path,
...)
}
)
}
if (p %in% stat.plots && is.norm) {
for (cnt in names(contrast.list)) {
disp(" Contrast: ",cnt)
samples <- names(unlist(contrast.list[[cnt]]))
mat <- as.matrix(object[,match(samples,colnames(object))])
switch(p,
deheatmap = {
files$deheatmap[[cnt]] <- diagplot.de.heatmap(mat,cnt,
output=output,path=path)
},
volcano = {
fc <- log2(make.fold.change(cnt,sample.list,object,1))
for (contrast in colnames(fc)) {
files$volcano[[contrast]] <- diagplot.volcano(
fc[,contrast],p.list[[cnt]],contrast,
fcut=thresholds$f,pcut=thresholds$p,
output=output,path=path)
}
},
biodist = {
files$biodist[[cnt]] <- diagplot.noiseq(object,
sample.list,covars,which.plot=p,output=output,
biodist.opts=list(p=p.list[[cnt]],
pcut=thresholds$p,name=cnt),path=path,...)
}
)
}
}
if (p %in% other.plots) {
switch(p,
filtered = {
files$filtered <- diagplot.filtered(object,annotation,
output=output,path=path)
}
)
}
if (p %in% venn.plots) {
switch(p,
venn = {
for (cnt in names(contrast.list)) {
disp(" Contrast: ",cnt)
if (!is.null(annotation)) {
alt.names <- as.character(annotation$gene_name)
names(alt.names) <- rownames(annotation)
}
else
alt.names <- NULL
files$venn[[cnt]] <- diagplot.venn(p.list[[cnt]],
pcut=thresholds$p,nam=cnt,output=output,path=path,
alt.names=alt.names)
}
}
)
}
}
return(files)
}
#' Boxplots wrapper for the metaseqR package
#'
#' A wrapper over the general boxplot function, suitable for matrices produced
#' and processed with the metaseqr package. Intended for internal use but can be
#' easily used as stand-alone. It can colors boxes based on group depending on
#' the name argument.
#'
#' @param mat the count data matrix.
#' @param name the names of the samples plotted on the boxplot. If \code{NULL},
#' the function check the column names of mat. If they are also \code{NULL}, sample
#' names are autogenerated. If \code{name="none"}, no sample names are plotted.
#' If name is a list, it should be the sample.list argument provided to the manin
#' metaseqr function. In that case, the boxes are colored per group.
#' @param log.it whether to log transform the values of mat or not. It can be
#' \code{TRUE}, \code{FALSE} or \code{"auto"} for auto-detection. Auto-detection
#' log transforms by default so that the boxplots are smooth and visible.
#' @param y.lim custom y-axis limits. Leave the string \code{"default"} for default
#' behavior.
#' @param is.norm a logical indicating whether object contains raw or normalized
#' data. It is not essential and it serves only plot annotation purposes.
#' @param output one or more R plotting device to direct the plot result to.
#' Supported mechanisms: \code{"x11"} (default), \code{"png"}, \code{"jpg"},
#' \code{"bmp"}, \code{"pdf"}, \code{"ps"} or \code{"json"}. The latter is
#' currently available for the creation of interactive volcano plots only when
#' reporting the output, through the highcharts javascript library (JSON for
#' boxplots not yet available).
#' @param path the path to create output files.
#' @param alt.names an optional vector of names, e.g. HUGO gene symbols, alternative
#' or complementary to the unique names of \code{f} or \code{p} (one of them must
#' be named!). It is used only in JSON output.
#' @param ... further arguments to be passed to plot devices, such as parameter
#' from \code{\link{par}}.
#' @return The filename of the boxplot produced if it's a file.
#' @export
#' @author Panagiotis Moulos
#' @examples
#' \dontrun{
#' require(DESeq)
#' data.matrix <- counts(makeExampleCountDataSet())
#' sample.list <- list(A=c("A1","A2"),B=c("B1","B2","B3"))
#' diagplot.boxplot(data.matrix,sample.list)
#'
#' norm.args <- get.defaults("normalization","deseq")
#' object <- normalize.deseq(data.matrix,sample.list,norm.args)
#' diagplot.boxplot(object,sample.list)
#'}
diagplot.boxplot <- function(mat,name=NULL,log.it="auto",y.lim="default",
is.norm=FALSE,output="x11",path=NULL,alt.names=NULL,...) {
if (is.null(path)) path <- getwd()
if (is.norm)
status<- "normalized"
else
status<- "raw"
# Need to log?
if (log.it=="auto") {
if (diff(range(mat,na.rm=TRUE))>1000)
mat <- log2disp(mat)
}
else if (log.it=="yes")
mat <- log2disp(mat)
# Define the axis limits based on user input
if (!is.numeric(y.lim) && y.lim=="default") {
min.y <- floor(min(mat))
max.y <- ceiling(max(mat))
}
else if (is.numeric(y.lim)) {
min.y <- y.lim[1]
max.y <- y.lim[2]
}
grouped <- FALSE
if (is.null(name)) {
if (is.null(colnames(mat)))
nams <- paste("Sample",1:ncol(mat),sep=" ")
else
nams <- colnames(mat)
}
else if (length(name)==1 && name=="none")
nams <- rep("",ncol(mat))
else if (is.list(name)) { # Is sample.list
nams <- unlist(name)
grouped <- TRUE
}
cols <- c("red3","green3","blue2","gold","skyblue","orange3","burlywood",
"red","blue","green","orange","darkgrey","green4","black","pink",
"brown","magenta","yellowgreen","pink4","seagreen4","darkcyan")
if (grouped) {
tmp <- as.numeric(factor(as.class.vector(name)))
b.cols <- cols[tmp]
}
else b.cols <- cols
mat.list <- list()
for (i in 1:ncol(mat))
mat.list[[i]] <- mat[,i]
names(mat.list) <- nams
if (output != "json") {
fil <- file.path(path,paste("boxplot_",status,".",output,sep=""))
graphics.open(output,fil)
if (!is.numeric(y.lim) && y.lim=="default")
b <- boxplot(mat.list,names=nams,col=b.cols,las=2,main=paste(
"Boxplot ",status,sep=""),...)
else
b <- boxplot(mat.list,names=nams,col=b.cols,ylim=c(min.y,max.y),
las=2,main=paste("Boxplot ",status,sep=""),...)
graphics.close(output)
}
else {
# Create boxplot object
b <- boxplot(mat.list,plot=FALSE)
colnames(b$stat) <- nams
# Locate the outliers
o.list <- lapply(names(mat.list),function(x,M,b) {
v <- b[,x]
o <- which(M[[x]]<v[1] | M[[x]]>v[5])
if (length(o)>0)
return(M[[x]][o])
else
return(NULL)
},mat.list,b$stat)
# Create output object
obj <- list(
x=NULL,
y=NULL,
plot=b,
samples=name,
ylims=c(min.y,max.y),
xlims=NULL,
status=status,
pcut=NULL,
fcut=NULL,
altnames=alt.names,
user=o.list
)
json <- boxplotToJSON(obj)
fil <- file.path(path,paste("boxplot_",status,".json",sep=""))
disp("Writing ",fil)
write(json,fil)
}
return(fil)
}
#' Multi-Dimensinal Scale plots or RNA-Seq samples
#'
#' Creates a Multi-Dimensional Scale plot for the given samples based on the count
#' data matrix. MDS plots are very useful for quality control as you can easily
#' see of samples of the same groups are clustered together based on the whole
#' dataset.
#'
#' @param x the count data matrix.
#' @param sample.list the list containing condition names and the samples under
#' each condition.
#' @param method which correlation method to use. Same as the method parameter in
#' \code{\link{cor}} function.
#' @param log.it whether to log transform the values of x or not.
#' @param output one or more R plotting device to direct the plot result to.
#' Supported mechanisms: \code{"x11"} (default), \code{"png"}, \code{"jpg"},
#' \code{"bmp"}, \code{"pdf"}, \code{"ps"} or \code{"json"}. The latter is
#' currently available for the creation of interactive volcano plots only when
#' reporting the output, through the highcharts javascript library.
#' @param path the path to create output files.
#' @param ... further arguments to be passed to plot devices, such as parameter
#' from \code{\link{par}}.
#' @return The filename of the MDS plot produced if it's a file.
#' @export
#' @author Panagiotis Moulos
#' @examples
#' \dontrun{
#' require(DESeq)
#' data.matrix <- counts(makeExampleCountDataSet())
#' sample.list <- list(A=c("A1","A2"),B=c("B1","B2","B3"))
#' diagplot.mds(data.matrix,sample.list)
#'}
diagplot.mds <- function(x,sample.list,method="spearman",log.it=TRUE,
output="x11",path=NULL,...) {
if (is.null(path)) path <- getwd()
classes <- as.factor(as.class.vector(sample.list))
design <- as.numeric(classes)
colspace <- c("red","blue","yellowgreen","orange","aquamarine2",
"pink2","seagreen4","brown","purple","chocolate")
pchspace <- c(20,17,15,16,8,3,2,0,1,4)
if (ncol(x)<3) {
warnwrap("MDS plot cannot be created with less than 3 samples! ",
"Skipping...")
return(NULL)
}
if (log.it)
y <- nat2log(x,base=2)
else
y <- x
d <- as.dist(0.5*(1-cor(y,method=method)))
mds.obj <- cmdscale(d,eig=TRUE,k=2)
xr <- diff(range(min(mds.obj$points[,1]),max(mds.obj$points[,1])))
yr <- diff(range(min(mds.obj$points[,2]),max(mds.obj$points[,2])))
xlim <- c(min(mds.obj$points[,1])-xr/10,max(mds.obj$points[,1])+xr/10)
ylim <- c(min(mds.obj$points[,2])-yr/10,max(mds.obj$points[,2])+yr/10)
if (output!="json") {
fil <- file.path(path,paste("mds.",output,sep=""))
if (output %in% c("pdf","ps","x11"))
graphics.open(output,fil,width=9,height=7)
else
graphics.open(output,fil,width=1024,height=768)
plot(mds.obj$points[,1],mds.obj$points[,2],
col=colspace[1:length(levels(classes))][design],
pch=pchspace[1:length(levels(classes))][design],
xlim=xlim,ylim=ylim,
main="MDS plot",xlab="MDS 1",ylab="MDS 2",
cex=0.9,cex.lab=0.9,cex.axis=0.9,cex.main=0.9)
text(mds.obj$points[,1],mds.obj$points[,2],labels=colnames(x),pos=3,
cex=0.7)
grid()
graphics.close(output)
}
else {
# Create output object
xx <- mds.obj$points[,1]
yy <- mds.obj$points[,2]
names(xx) <- names(yy) <- unlist(sample.list)
obj <- list(
x=xx,
y=yy,
plot=NULL,
samples=sample.list,
ylim=ylim,
xlim=xlim,
status=NULL,
pcut=NULL,
fcut=NULL,
altnames=NULL,
user=NULL
)
json <- mdsToJSON(obj)
fil <- file.path(path,"mds.json")
disp("Writing ",fil)
write(json,fil)
}
return(fil)
}
#' Massive X-Y, M-D correlation plots
#'
#' This function uses the read counts matrix to create pairwise correlation plots.
#' The upper diagonal of the final image contains simple scatterplots of each
#' sample against each other (log2 scale) while the lower diagonal contains
#' mean-difference plots for the same samples (log2 scale). This type of diagnostic
#' plot may not be interpretable for more than 10 samples.
#'
#' @param x the read counts matrix or data frame.
#' @param output one or more R plotting device to direct the plot result to.
#' Supported mechanisms: \code{"x11"} (default), \code{"png"}, \code{"jpg"},
#' \code{"bmp"}, \code{"pdf"} or \code{"ps"}.
#' @param path the path to create output files.
#' @param ... further arguments to be passed to plot devices, such as parameter
#' from \code{\link{par}}.
#' @return The filename of the pairwise comparisons plot produced if it's a file.
#' @export
#' @author Panagiotis Moulos
#' @examples
#' \dontrun{
#' require(DESeq)
#' data.matrix <- counts(makeExampleCountDataSet())
#' diagplot.pairs(data.matrix)
#'}
diagplot.pairs <- function(x,output="x11",path=NULL,...) {
x <- as.matrix(x)
x <- nat2log(x)
n <- ncol(x)
if (!is.null(colnames(x)))
nams <- colnames(x)
else
nams <- paste("Sample_",1:ncol(x),sep="")
if (!is.null(path))
fil <- file.path(path,paste("correlation_pairs",output,sep="."))
else
fil <- paste("correlation_pairs",output,sep=".")
if (output %in% c("pdf","ps","x11"))
graphics.open(output,fil,width=12,height=12)
else {
if (ncol(x)<=5)
graphics.open(output,fil,width=800,height=800,res=100)
else
graphics.open(output,fil,width=1024,height=1024,res=150)
}
# Setup the grid
par(mfrow=c(n,n),mar=c(1,1,1,1),oma=c(1,1,0,0),mgp=c(2,0.5,0),cex.axis=0.6,
cex.lab=0.6)
# Plot
for (i in 1:n)
{
for (j in 1:n)
{
if (i==j)
{
plot(0:10,0:10,type="n",xaxt="n",yaxt="n",xlab="",ylab="") # Diagonal
text(c(3,5,3),c(9.5,5,1),c("X-Y plots",nams[i],"M-D plots"),
cex=c(0.8,1,0.8))
arrows(6,9.5,9.5,9.5,angle=20,length=0.1,lwd=0.8,cex=0.8)
arrows(0.2,3.2,0.2,0.2,angle=20,length=0.1,lwd=0.8,cex=0.8)
}
else if (i<j) # XY plot
{
plot(x[,i],x[,j],pch=20,col="blue",cex=0.4,xlab=nams[i],
ylab=nams[j],...)
lines(lowess(x[,i],x[,j]),col="red")
cc <- paste("cor:",formatC(cor(x[,i],x[,j]),digits=3))
text(3,max(x[,j]-1),labels=cc,cex=0.7,)
#grid()
}
else if (i>j) # MD plot
{
plot((x[,i]+x[,j])/2,x[,j]-x[,i],pch=20,col="blue",cex=0.4,...)
lines(lowess((x[,i]+x[,j])/2,x[,j]-x[,i]),col="red")
#grid()
}
}
}
graphics.close(output)
return(fil)
}
#' Summarized correlation plots
#'
#' This function uses the read counts matrix to create heatmap or correlogram
#' correlation plots.
#'
#' @param mat the read counts matrix or data frame.
#' @param type create heatmap of correlogram plots.
#' @param output one or more R plotting device to direct the plot result to.
#' Supported mechanisms: \code{"x11"} (default), \code{"png"}, \code{"jpg"},
#' \code{"bmp"}, \code{"pdf"} or \code{"ps"}.
#' @param path the path to create output files.
#' @param ... further arguments to be passed to plot devices, such as parameter
#' from \code{\link{par}}.
#' @return The filename of the pairwise comparisons plot produced if it's a file.
#' @export
#' @author Panagiotis Moulos
#' @examples
#' \dontrun{
#' require(DESeq)
#' data.matrix <- counts(makeExampleCountDataSet())
#' diagplot.cor(data.matrix,type="heatmap")
#' diagplot.cor(data.matrix,type="correlogram")
#'}
diagplot.cor <- function(mat,type=c("heatmap","correlogram"),output="x11",
path=NULL,...) {
x <- as.matrix(mat)
type <- tolower(type[1])
check.text.args("type",type,c("heatmap","correlogram"))
#if (!require(corrplot) && type=="correlogram")
# stop("R package corrplot is required!")
cor.mat <- cor(mat)
if (!is.null(colnames(mat)))
colnames(cor.mat) <- colnames(mat)
if (!is.null(path))
fil <- file.path(path,paste("correlation_",type,".",output,sep=""))
else
fil <- paste("correlation_",type,".",output,sep="")
if (output %in% c("pdf","ps","x11"))
graphics.open(output,fil,width=7,height=7)
else
graphics.open(output,fil,width=640,height=640,res=100)
if (type=="correlogram")
corrplot(cor.mat,method="ellipse",order="hclust",...)
else if (type=="heatmap") {
n <- dim(cor.mat)[1]
labs <- matrix(NA,n,n)
for (i in 1:n)
for (j in 1:n)
labs[i,j] <- sprintf("%.2f",cor.mat[i,j])
if (n <= 5)
notecex <- 1.2
else if (n > 5 & n < 10)
notecex <- 0.9
else
notecex <- 0.7
heatmap.2(cor.mat,col=colorRampPalette(c("yellow","grey","blue")),
revC=TRUE,trace="none",symm=TRUE,Colv=TRUE,cellnote=labs,
keysize=1,density.info="density",notecex=notecex,cexCol=0.9,
cexRow=0.9,font.lab=2)
}
graphics.close(output)
return(fil)
}
#' Diagnostic plots based on the EDASeq package
#'
#' A wrapper around the plotting functions availale in the EDASeq normalization
#' Bioconductor package. For analytical explanation of each plot please see the
#' vignette of the EDASeq package. It is best to use this function through the
#' main plotting function \code{\link{diagplot.metaseqr}}.
#'
#' @param x the count data matrix.
#' @param sample.list the list containing condition names and the samples under
#' each condition.
#' @param covar The covariate to plot counts against. Usually \code{"gc"} or
#' \code{"length"}.
#' @param is.norm a logical indicating whether object contains raw or normalized
#' data. It is not essential and it serves only plot annotation purposes.
#' @param which.plot the EDASeq package plot to generate. It can be one or more
#' of \code{"meanvar"}, \code{"meandiff"}, \code{"gcbias"} or \code{"lengthbias"}.
#' Please refer to the documentation of the EDASeq package for details on the use
#' of these plots. The \code{which.plot="lengthbias"} case is not covered by
#' EDASeq documentation, however it is similar to the GC-bias plot when the
#' covariate is the gene length instead of the GC content.
#' @param output one or more R plotting device to direct the plot result to.
#' Supported mechanisms: \code{"x11"} (default), \code{"png"}, \code{"jpg"},
#' \code{"bmp"}, \code{"pdf"} or \code{"ps"}.
#' @param path the path to create output files.
#' @param ... further arguments to be passed to plot devices, such as parameter
#' from \code{\link{par}}.
#' @return The filenames of the plot produced in a named list with names the
#' which.plot argument. If \code{output="x11"}, no output filenames are produced.
#' @export
#' @author Panagiotis Moulos
#' @examples
#' \dontrun{
#' require(DESeq)
#' data.matrix <- counts(makeExampleCountDataSet())
#' sample.list <- list(A=c("A1","A2"),B=c("B1","B2","B3"))
#' gc <- runif(nrow(data.matrix))
#' diagplot.edaseq(data.matrix,sample.list,which.plot="meandiff")
#'}
diagplot.edaseq <- function(x,sample.list,covar=NULL,is.norm=FALSE,
which.plot=c("meanvar","meandiff","gcbias","lengthbias"),output="x11",
path=NULL,...) {
if (is.null(path)) path <- getwd()
check.text.args("which.plot",which.plot,c("meanvar","meandiff","gcbias",
"lengthbias"),multiarg=TRUE)
if (is.null(covar) && which.plot %in% c("gcbias","lengthbias"))
stopwrap("\"covar\" argument is required when \"which.plot\" is ",
"\"gcbias\ or \"lengthbias\"!")
if (is.norm)
status <- "normalized"
else
status <- "raw"
if (is.null(covar)) covar <- rep(NA,nrow(x))
s <- newSeqExpressionSet(x,phenoData=AnnotatedDataFrame(
data.frame(conditions=factor(as.class.vector(sample.list)),
row.names=colnames(x))),featureData=AnnotatedDataFrame(data.frame(
gc=covar,length=covar,row.names=rownames(x))))
switch(which.plot,
meandiff = {
fil <- vector("list",length(sample.list))
names(fil) <- names(sample.list)
for (n in names(sample.list)) {
if (length(sample.list[[n]])==1) {
warnwrap("Cannot create a mean-difference plot with one ",
"sample per condition! Skipping...")
next
}
pair.matrix <- combn(1:length(sample.list[[n]]),2)
fil[[n]] <- vector("list",ncol(pair.matrix))
for (i in 1:ncol(pair.matrix)) {
s1 <- sample.list[[n]][pair.matrix[1,i]]
s2 <- sample.list[[n]][pair.matrix[2,i]]
fil[[n]][[i]] <- file.path(path,paste(which.plot,"_",
status,"_",n,"_",s1,"_",s2,".",output,sep=""))
names(fil[[n]][i]) <- paste(s1,"vs",s2,sep="_")
graphics.open(output,fil[[n]][[i]])
MDPlot(s,y=pair.matrix[,i],main=paste("MD plot for ",n," ",
status," samples ",s1," and ",s2,sep=""),cex.main=0.9)
graphics.close(output)
}
}
},
meanvar = {
fil <- vector("list",length(sample.list))
names(fil) <- names(sample.list)
for (n in names(sample.list)) {
if (length(sample.list[[n]])==1) {
warnwrap("Cannot create a mean-variance plot with one ",
"sample per condition! Skipping...")
next
}
pair.matrix <- combn(1:length(sample.list[[n]]),2)
fil[[n]] <- vector("list",ncol(pair.matrix))
for (i in 1:ncol(pair.matrix)) {
s1 <- sample.list[[n]][pair.matrix[1,i]]
s2 <- sample.list[[n]][pair.matrix[2,i]]
fil[[n]][[i]] <- file.path(path,paste(which.plot,"_",status,
"_",n,"_",s1,"_",s2,".",output,sep=""))
names(fil[[n]][i]) <- paste(s1,"vs",s2,sep="_")
graphics.open(output,fil[[n]][[i]])
suppressWarnings(meanVarPlot(s,main=paste("MV plot for ",n,
" ",status," samples ",s1," and ",s2,sep=""),
cex.main=0.9))
graphics.close(output)
}
}
},
gcbias = {
if (!output=="json") {
fil <- file.path(path,paste(which.plot,"_",status,".",output,
sep=""))
graphics.open(output,fil)
biasPlot(s,"gc",xlim=c(0.1,0.9),log=TRUE,ylim=c(0,15),
main=paste("Expression - GC content ",status,sep=""))
grid()
graphics.close(output)
}
else {
obj <- list(
x=NULL,
y=NULL,
plot=NULL,
samples=sample.list,
ylim=NULL,
xlim=NULL,
status=status,
pcut=NULL,
fcut=NULL,
altnames=NULL,
user=list(counts=x,covar=covar,covarname="GC content")
)
json <- biasPlotToJSON(obj)
fil <- file.path(path,paste(which.plot,"_",status,".json",
sep=""))
disp("Writing ",fil)
write(json,fil)
}
},
lengthbias = {
if (output!="json") {
fil <- file.path(path,paste(which.plot,"_",status,".",output,
sep=""))
graphics.open(output,fil)
biasPlot(s,"length",log=TRUE,ylim=c(0,10),
main=paste("Expression - Gene length ",status,sep=""))
grid()
graphics.close(output)
}
else {
obj <- list(
x=NULL,
y=NULL,
plot=NULL,
samples=sample.list,
ylim=NULL,
xlim=NULL,
status=status,
pcut=NULL,
fcut=NULL,
altnames=NULL,
user=list(counts=x,covar=covar,
covarname="Gene/transcript length")
)
json <- biasPlotToJSON(obj)
fil <- file.path(path,paste(which.plot,"_",status,".json",
sep=""))
disp("Writing ",fil)
write(json,fil)
}
}
)
return(fil)
}
#' Diagnostic plots based on the NOISeq package
#'
#' A wrapper around the plotting functions availale in the NOISeq Bioconductor
#' package. For analytical explanation of each plot please see the vignette of
#' the NOISeq package. It is best to use this function through the main plotting
#' function \code{\link{diagplot.metaseqr}}.
#'
#' @param x the count data matrix.
#' @param sample.list the list containing condition names and the samples under
#' each condition.
#' @param covars a list (whose annotation elements are ideally a subset of an
#' annotation data frame produced by \code{\link{get.annotation}})
#' with the following members: data (the data matrix), length (gene length), gc
#' (the gene gc_content), chromosome (a data frame with chromosome name and
#' co-ordinates), factors (a factor with the experimental condition names
#' replicated by the number of samples in each experimental condition) and biotype
#' (each gene's biotype as depicted in Ensembl-like annotations).
#' @param which.plot the NOISeq package plot to generate. It can be one or more
#' of \code{"biodetection"}, \code{"countsbio"}, \code{"saturation"},
#' \code{"rnacomp"}, \code{"readnoise"} or \code{"biodist"}. Please refer to the
#' documentation of the EDASeq package for details on the use of these plots. The
#' \code{which.plot="saturation"} case is modified to be more informative by
#' producing two kinds of plots. See \code{\link{diagplot.noiseq.saturation}}.
#' @param biodist.opts a list with the following members: p (a vector of p-values,
#' e.g. the p-values of a contrast), pcut (a unique number depicting a p-value
#' cutoff, required for the \code{"biodist"} case), name (a name for the
#' \code{"biodist"} plot, e.g. the name of the contrast.
#' @param output one or more R plotting device to direct the plot result to.
#' Supported mechanisms: \code{"x11"} (default), \code{"png"}, \code{"jpg"},
#' \code{"bmp"}, \code{"pdf"} or \code{"ps"}.
#' @param path the path to create output files.
#' @param is.norm a logical indicating whether object contains raw or normalized
#' data. It is not essential and it serves only plot annotation purposes.
#' @param ... further arguments to be passed to plot devices, such as parameter
#' from \code{\link{par}}.
#' @return The filenames of the plots produced in a named list with names the
#' \code{which.plot} argument. If \code{output="x11"}, no output filenames are
#' produced.
#' @note Please note that in case of \code{"biodist"} plots, the behavior of the
#' function is unstable, mostly due to the very specific inputs this plotting
#' function accepts in the NOISeq package. We have tried to predict unstable
#' behavior and avoid exceptions through the use of tryCatch but it's still
#' possible that you might run onto an error.
#' @export
#' @author Panagiotis Moulos
#' @examples
#' \dontrun{
#' require(DESeq)
#' data.matrix <- counts(makeExampleCountDataSet())
#' sample.list <- list(A=c("A1","A2"),B=c("B1","B2","B3"))
#' lengths <- round(1000*runif(nrow(data.matrix)))
#' starts <- round(1000*runif(nrow(data.matrix)))
#' ends <- starts + lengths
#' covars <- list(
#' data=data.matrix,
#' length=lengths,
#' gc=runif(nrow(data.matrix)),
#' chromosome=data.frame(
#' chromosome=c(rep("chr1",nrow(data.matrix)/2),rep("chr2",nrow(data.matrix)/2)),
#' start=starts,
#' end=ends
#' ),
#' factors=data.frame(class=as.class.vector(sample.list)),
#' biotype=c(rep("protein_coding",nrow(data.matrix)/2),rep("ncRNA",
#' nrow(data.matrix)/2))
#' )
#' p <- runif(nrow(data.matrix))
#' diagplot.noiseq(data.matrix,sample.list,covars=covars,
#' biodist.opts=list(p=p,pcut=0.1,name="A_vs_B"))
#'}
diagplot.noiseq <- function(x,sample.list,covars,which.plot=c("biodetection",
"countsbio","saturation","rnacomp","readnoise","biodist"),output="x11",
biodist.opts=list(p=NULL,pcut=NULL,name=NULL),path=NULL,is.norm=FALSE,
...) {
if (is.null(path)) path <- getwd()
# covars is a list of gc-content, factors, length, biotype, chromosomes,
# factors, basically copy of the noiseq object
which.plot <- tolower(which.plot[1])
check.text.args("which.plot",which.plot,c("biodetection","countsbio",
"saturation","readnoise","rnacomp","biodist"),multiarg=FALSE)
if (missing(covars))
stopwrap("\"covars\" argument is required with NOISeq specific plots!")
else {
covars$biotype <- as.character(covars$biotype)
names(covars$length) <- names(covars$gc) <-
rownames(covars$chromosome) <- names(covars$biotype) <-
rownames(x)
}
if (which.plot=="biodist") {
if (is.null(biodist.opts$p))
stopwrap("A p-value must be provided for the \"biodist\" plot!")
if (is.null(biodist.opts$pcut) || is.na(biodist.opts$pcut))
biodist.opts$pcut=0.05
}
if (is.norm)
status<- "normalized"
else
status<- "raw"
# All of these plots are NOISeq specific so we need a local NOISeq object
if (any(is.na(unique(covars$biotype))))
covars$biotype=NULL # Otherwise, it will probably crash
local.obj <- NOISeq::readData(
data=x,
length=covars$gene.length,
gc=covars$gc.content,
chromosome=covars$chromosome,
#factors=data.frame(class=covars$factors),
factors=covars$factors,
biotype=covars$biotype
)
switch(which.plot,
biodetection = {
diagplot.data <- NOISeq::dat(local.obj,type=which.plot)
samples <- unlist(sample.list)
if (output!="json") {
fil <- character(length(samples))
names(fil) <- samples
for (i in 1:length(samples)) {
fil[samples[i]] <- file.path(path,paste(which.plot,"_",
samples[i],".",output,sep=""))
if (output %in% c("pdf","ps","x11"))
graphics.open(output,fil[samples[i]],width=9,height=7)
else
graphics.open(output,fil[samples[i]],width=1024,height=768)
explo.plot(diagplot.data,samples=i)
graphics.close(output)
}
}
else {
diagplot.data.save = NOISeq::dat2save(diagplot.data)
obj <- list(
x=NULL,
y=NULL,
plot=NULL,
samples=sample.list,
ylims=NULL,
xlims=NULL,
status=status,
pcut=NULL,
fcut=NULL,
altnames=covars$gene_name,
user=list(plotdata=diagplot.data.save,covars=covars)
)
json <- bioDetectionToJSON(obj)
fil <- character(length(samples))
names(fil) <- samples
for (i in 1:length(samples)) {
fil[samples[i]] <- file.path(path,
paste(which.plot,"_",samples[i],".json",sep=""))
disp("Writing ",fil[samples[i]])
write(json[[i]],fil[samples[i]])
}
}
},
countsbio = {
samples <- unlist(sample.list)
if (output!="json") {
diagplot.data <- NOISeq::dat(local.obj,type=which.plot,
factor=NULL)
fil <- character(length(samples))
names(fil) <- samples
for (i in 1:length(samples)) {
fil[samples[i]] <- file.path(path,paste(which.plot,"_",
samples[i],".",output,sep=""))
if (output %in% c("pdf","ps","x11"))
graphics.open(output,fil[samples[i]],width=9,height=7)
else
graphics.open(output,fil[samples[i]],width=1024,
height=768)
explo.plot(diagplot.data,samples=i,plottype="boxplot")
graphics.close(output)
}
}
else {
colnames(x) <- unlist(sample.list)
obj <- list(
x=NULL,
y=NULL,
plot=NULL,
samples=sample.list,
ylims=NULL,
xlims=NULL,
status=status,
pcut=NULL,
fcut=NULL,
altnames=covars$gene_name,
user=list(counts=nat2log(x),covars=covars)
)
# Write JSON by sample
fil <- vector("list",2)
names(fil) <- c("sample","biotype")
fil[["sample"]] <- character(length(samples))
names(fil[["sample"]]) <- samples
bts <- unique(as.character(obj$user$covars$biotype))
fil[["biotype"]] <- character(length(bts))
names(fil[["biotype"]]) <- bts
json <- countsBioToJSON(obj,by="sample")
for (i in 1:length(samples)) {
fil[["sample"]][samples[i]] <- file.path(path,
paste(which.plot,"_",samples[i],".json",sep=""))
disp("Writing ",fil[["sample"]][samples[i]])
write(json[[i]],fil[["sample"]][samples[i]])
}
json <- countsBioToJSON(obj,by="biotype")
for (i in 1:length(bts)) {
fil[["biotype"]][bts[i]] <- file.path(path,
paste(which.plot,"_",bts[i],".json",sep=""))
disp("Writing ",fil[["biotype"]][bts[i]])
write(json[[i]],fil[["biotype"]][bts[i]])
}
}
},
saturation = {
# For 10 saturation points
diagplot.data <- NOISeq::dat(local.obj,k=0,ndepth=9,type=which.plot)
d2s <- NOISeq::dat2save(diagplot.data)
if (output != "json")
fil <- diagplot.noiseq.saturation(d2s,output,covars$biotype,
path=path)
else {
samples <- unlist(sample.list)
obj <- list(
x=NULL,
y=NULL,
plot=NULL,
samples=sample.list,
ylims=NULL,
xlims=NULL,
status=status,
pcut=NULL,
fcut=NULL,
altnames=covars$gene_name,
user=list(plotdata=d2s)
)
# Write JSON by sample
fil <- vector("list",2)
names(fil) <- c("sample","biotype")
fil[["sample"]] <- character(length(samples))
names(fil[["sample"]]) <- samples
json <- bioSaturationToJSON(obj,by="sample")
for (i in 1:length(samples)) {
fil[["sample"]][samples[i]] <- file.path(path,
paste(which.plot,"_",samples[i],".json",sep=""))
disp("Writing ",fil[["sample"]][samples[i]])
write(json[[i]],fil[["sample"]][samples[i]])
}
json <- bioSaturationToJSON(obj,by="biotype")
fil[["biotype"]] <- character(length(json))
names(fil[["biotype"]]) <- names(json)
for (n in names(json)) {
fil[["biotype"]][n] <- file.path(path,
paste(which.plot,"_",n,".json",sep=""))
disp("Writing ",fil[["biotype"]][n])
write(json[[n]],fil[["biotype"]][n])
}
}
},
rnacomp = {
if (ncol(local.obj)<3) {
warnwrap("RNA composition plot cannot be created with less ",
"than 3 samples! Skipping...")
return(NULL)
}
if (ncol(local.obj)>12) {
warnwrap("RNA composition plot cannot be created with more ",
"than 12 samples! Skipping...")
return(NULL)
}
diagplot.data <- NOISeq::dat(local.obj,type="cd")
fil <- file.path(path,paste(which.plot,"_",status,".",output,
sep=""))
graphics.open(output,fil)
explo.plot(diagplot.data)
grid()
graphics.close(output)
},
readnoise = {
D <- cddat(local.obj)
if (output!="json") {
fil <- file.path(path,paste(which.plot,".",output,sep=""))
graphics.open(output,fil)
cdplot(D,main="RNA-Seq reads noise")
grid()
graphics.close(output)
}
else {
colnames(D$data2plot)[2:ncol(D$data2plot)] <-
unlist(sample.list)
obj <- list(
x=NULL,
y=NULL,
plot=NULL,
samples=sample.list,
xlim=NULL,
ylim=NULL,
status=NULL,
pcut=NULL,
fcut=NULL,
altnames=NULL,
user=D$data2plot
)
json <- readNoiseToJSON(obj)
fil <- file.path(path,paste(which.plot,".json",sep=""))
disp("Writing ",fil)
write(json,fil)
}
},
biodist = { # We have to fake a noiseq object
p <- biodist.opts$p
if (is.matrix(p)) p <- p[,1]
dummy <- new("Output",
comparison=c("Dummy.1","Dummy.2"),
factor=c("class"),
k=1,
lc=1,
method="n",
replicates="biological",
results=list(
data.frame(
Dummy.1=rep(1,length(p)),
Dummy.2=rep(1,length(p)),
M=rep(1,length(p)),
D=rep(1,length(p)),
prob=as.numeric(p),
ranking=rep(1,length(p)),
Length=rep(1,length(p)),
GC=rep(1,length(p)),
Chrom=as.character(covars$chromosome[,1]),
GeneStart=covars$chromosome[,2],
GeneEnd=covars$chromosome[,3],
Biotype=covars$biotype
)
),
nss=5,
pnr=0.2,
v=0.02
)
if (!is.null(biodist.opts$name))
fil <- file.path(path,paste(which.plot,"_",biodist.opts$name,
".",output,sep=""))
else
fil <- file.path(path,paste(which.plot,".",output,sep=""))
if (output %in% c("pdf","ps","x11"))
graphics.open(output,fil,width=10,height=6)
else
graphics.open(output,fil,width=1024,height=640)
tryCatch( # A lot of times, there is a problem with this function
DE.plot(dummy,chromosomes=NULL,q=biodist.opts$pcut,
graphic="distr"),
error=function(e) {
disp(" Known problem with NOISeq and external ",
"p-values detected! Trying to make a plot with ",
"alternative p-values (median of p-value ",
"distribution)...")
fil="error"
tryCatch(
DE.plot(dummy,chromosomes=NULL,
q=quantile(biodist.opts$p,0.5),
graphic="distr"),
error=function(e) {
disp(" Cannot create DEG biotype plot! This ",
"is not related to a problem with the ",
"results. Excluding...")
fil="error"
},
finally=""
)
},
finally=""
)
graphics.close(output)
}
)
return(fil)
}
#' Simpler implementation of saturation plots inspired from NOISeq package
#'
#' Helper function for \code{\link{diagplot.noiseq}} to plot feature detection
#' saturation as presented in the NOISeq package vignette. It has two main outputs:
#' a set of figures, one for each input sample depicting the saturation for each
#' biotype and one single multiplot which depicts the saturation of all samples
#' for each biotype. It expands the saturation plots of NOISeq by allowing more
#' samples to be examined in a simpler way. Don't use this function directly. Use
#' either \code{\link{diagplot.metaseqr}} or \code{\link{diagplot.noiseq}}.
#'
#' @param x the count data matrix.
#' @param o one or more R plotting device to direct the plot result to. Supported
#' mechanisms: \code{"x11"} (default), \code{"png"}, \code{"jpg"}, \code{"bmp"},
#' \code{"pdf"} or \code{"ps"}.
#' @param tb the vector of biotypes, one for each row of x.
#' @param path the path to create output files.
#' @return The filenames of the plots produced in a named list with names the
#' \code{which.plot} argument. If \code{output="x11"}, no output filenames are
#' produced.
#' @author Panagiotis Moulos
#' @examples
#' \dontrun{
#' require(DESeq)
#' data.matrix <- counts(makeExampleCountDataSet())
#' biotype=c(rep("protein_coding",nrow(data.matrix)/2),rep("ncRNA",
#' nrow(data.matrix)/2))
#' diagplot.noiseq.saturation(data.matrix,"x11",biotype)
#'}
diagplot.noiseq.saturation <- function(x,o,tb,path=NULL) {
if (is.null(path)) path <- getwd()
if (length(unique(tb))==1) {
warnwrap("Saturation plot cannot be created with only one biotype! ",
"Skipping...")
return(NULL)
}
total.biotypes <- table(tb)
the.biotypes <- names(tb)
biotypes <- colnames(x[[1]][,2:ncol(x[[1]])])
colspace <- c("red3","green4","blue2","orange3","burlywood",
"lightpink4","gold","skyblue","red2","green2","firebrick3",
"orange4","yellow4","skyblue3","tan4","gray40",
"brown2","darkgoldenrod","cyan3","coral2","cadetblue",
"bisque3","blueviolet","chocolate3","darkkhaki","dodgerblue")
pchspace <- c(rep(c(15,16,17,18),6),15)
# Plot all biotypes per sample
f.sample <- character(length(names(x)))
names(f.sample) <- names(x)
for (n in names(x)) {
f.sample[n] <- file.path(path,paste("saturation_",n,".",o,sep=""))
if (o %in% c("pdf","ps","x11"))
graphics.open(o,f.sample[n],width=10,height=7)
else
graphics.open(o,f.sample[n],width=1024,height=800)
y <- x[[n]]
sep <- match(c("global","protein_coding"),colnames(y))
yab <- cbind(y[,"depth"],y[,sep])
ynab <- y[,-sep]
colnames(yab)[1] <- colnames(ynab)[1] <- "depth"
xlim <- range(y[,"depth"])
ylim.ab <- range(yab[,2:ncol(yab)])
ylim.nab <- range(ynab[,2:ncol(ynab)])
par(cex.axis=0.9,cex.main=1,cex.lab=0.9,font.lab=2,font.axis=2,pty="m",
lty=2,lwd=1.5,mfrow=c(1,2))
plot.new()
plot.window(xlim,ylim.nab)
axis(1,at=pretty(xlim,10),labels=as.character(pretty(xlim,10)/1e+6))
axis(2,at=pretty(ylim.nab,10))
title(main="Non abundant biotype detection saturation",
xlab="Depth in millions of reads",ylab="Detected features")
co <- 0
for (b in biotypes) {
co <- co + 1
if (b=="global" || b=="protein_coding") {
# Silently do nothing
}
else {
lines(ynab[,"depth"],ynab[,b],col=colspace[co])
points(ynab[,"depth"],ynab[,b],pch=pchspace[co],
col=colspace[co],cex=1)
}
}
grid()
graphics::legend(
x="topleft",legend=colnames(ynab)[2:ncol(ynab)],xjust=1,yjust=0,
box.lty=0,x.intersp=0.5,cex=0.6,text.font=2,
col=colspace[1:(ncol(ynab)-1)],pch=pchspace[1:(ncol(ynab)-1)]
)
plot.new()
plot.window(xlim,ylim.ab)
axis(1,at=pretty(xlim,10),labels=as.character(pretty(xlim,10)/1e+6))
axis(2,at=pretty(ylim.ab,10))
title(main="Abundant biotype detection saturation",
xlab="Depth in millions of reads",ylab="Detected features")
co <- 0
for (b in c("global","protein_coding")) {
co <- co + 1
lines(yab[,"depth"],yab[,b],col=colspace[co])
points(yab[,"depth"],yab[,b],pch=16,col=colspace[co],cex=1.2)
}
grid()
graphics::legend(
x="topleft",legend=c("global","protein_coding"),xjust=1,yjust=0,
box.lty=0,lty=2,x.intersp=0.5,cex=0.7,text.font=2,
col=colspace[1:2],pch=pchspace[1:2]
)
mtext(n,side=3,line=-1.5,outer=TRUE,font=2,cex=1.3)
graphics.close(o)
}
# Plot all samples per biotype
g <- make.grid(length(biotypes))
f.all <- file.path(path,paste("biotype_saturation.",o,sep=""))
if (o %in% c("pdf","ps"))
graphics.open(o,f.all,width=14,height=14)
else
graphics.open(o,f.all,width=1600,height=1600,res=150)
par(cex.axis=0.8,cex.main=0.9,cex.lab=0.8,pty="m",lty=2,lwd=1.5,mfrow=g,
mar=c(3,3,1,1),oma=c(1,1,0,0),mgp=c(2,0.5,0))
for (b in biotypes) {
y <- depth <- vector("list",length(x))
names(y) <- names(depth) <- names(x)
for (n in names(x)) {
y[[n]] <- x[[n]][,b]
depth[[n]] <- x[[n]][,"depth"]
}
y <- do.call("cbind",y)
xlim <- range(do.call("c",depth))
ylim <- range(y)
plot.new()
plot.window(xlim,ylim)
axis(1,at=pretty(xlim,5),labels=as.character(pretty(xlim,5)/1e+6),
line=0.5)
axis(2,at=pretty(ylim,5),line=0.5)
title(main=b,xlab="Depth in millions of reads",
ylab="Detected features")
co <- 0
for (n in colnames(y)) {
co <- co + 1
lines(depth[[n]],y[,n],col=colspace[co])
points(depth[[n]],y[,n],pch=pchspace[co],col=colspace[co])
}
grid()
graphics::legend(
x="bottomright",legend=colnames(y),xjust=1,yjust=0,
box.lty=0,x.intersp=0.5,
col=colspace[1:length(colnames(y))],
pch=pchspace[1:length(colnames(y))]
)
}
graphics.close(o)
return(list(sample=f.sample,biotype=f.all))
}
#' (Interactive) volcano plots of differentially expressed genes
#'
#' This function plots a volcano plot or returns a JSON string which is used to
#' render aninteractive in case of HTML reporting.
#'
#' @param f the fold changes which are to be plotted on the x-axis.
#' @param p the p-values whose -log10 transformation is going to be plotted on
#' the y-axis.
#' @param con an optional string depicting a name (e.g. the contrast name) to
#' appear in the title of the volcano diagplot.
#' @param fcut a fold change cutoff so as to draw two vertical lines indicating
#' the cutoff threshold for biological significance.
#' @param pcut a p-value cutoff so as to draw a horizontal line indicating the
#' cutoff threshold for statistical significance.
#' @param alt.names an optional vector of names, e.g. HUGO gene symbols, alternative
#' or complementary to the unique names of \code{f} or \code{p} (one of them must
#' be named!). It is used only in JSON output.
#' @param output one or more R plotting device to direct the plot result to.
#' Supported mechanisms: \code{"x11"} (default), \code{"png"}, \code{"jpg"},
#' \code{"bmp"}, \code{"pdf"}, \code{"ps"} or \code{"json"}. The latter is currently
#' available for the creation of interactive volcano plots only when reporting the
#' output, through the highcharts javascript library.
#' @param path the path to create output files.
#' @param ... further arguments to be passed to plot devices, such as parameter
#' from \code{\link{par}}.
#' @return The filenames of the plots produced in a named list with names the
#' \code{which.plot} argument. If \code{output="x11"}, no output filenames are
#' produced.
#' @export
#' @author Panagiotis Moulos
#' @examples
#' \dontrun{
#' require(DESeq)
#' data.matrix <- counts(makeExampleCountDataSet())
#' sample.list <- list(A=c("A1","A2"),B=c("B1","B2","B3"))
#' contrast <- "A_vs_B"
#' M <- norm.edger(data.matrix,sample.list)
#' p <- stat.edger(M,sample.list,contrast)
#' ma <- apply(M[,sample.list$A],1,mean)
#' mb <- apply(M[,sample.list$B],1,mean)
#' f <- log2(ifelse(mb==0,1,mb)/ifelse(ma==0,1,ma))
#' diagplot.volcano(f,p,con=contrast)
#' j <- diagplot.volcano(f,p,con=contrast,output="json")
#'}
diagplot.volcano <- function(f,p,con=NULL,fcut=1,pcut=0.05,alt.names=NULL,
output="x11",path=NULL,...) { # output can be json here...
## Check rjson
#if ("json" %in% output && !require(rjson))
# stopwrap("R package rjson is required to create interactive volcano plot!")
if (is.null(path)) path <- getwd()
if (is.null(con))
con <- conn <- ""
else {
conn <- con
con <- paste("for ",con)
}
fil <- file.path(path,paste("volcano_plot_",conn,".",output,sep=""))
if (output!="json") {
if (output %in% c("pdf","ps","x11"))
graphics.open(output,fil,width=8,height=10)
else
graphics.open(output,fil,width=768,height=1024,res=100)
}
rem <- which(is.na(p))
if (length(rem)>0) {
p <- p[-rem]
f <- f[-rem]
if (!is.null(alt.names))
alt.names <- alt.names[-rem]
}
# Fix problem with extremely low p-values, only for display purposes though
p.zero <- which(p==0)
if (length(p.zero)>0)
p[p.zero] <- runif(length(p.zero),0,1e-256)
xlim <- c(-max(abs(f)),max(abs(f)))
ylim <- c(0,ceiling(-log10(min(p))))
up <- which(f>=fcut & p<pcut)
down <- which(f<=-fcut & p<pcut)
u <- union(up,down)
alt.names.neutral <- NULL
if (length(u)>0) {
ff <- f[-u]
pp <- p[-u]
if (!is.null(alt.names))
alt.names.neutral <- alt.names[-u]
}
else {
ff <- f
pp <- p
if (!is.null(alt.names))
alt.names.neutral <- alt.names
}
if (output!="json") {
par(cex.main=1.1,cex.lab=1.1,cex.axis=1.1,font.lab=2,font.axis=2,
pty="m",lwd=1.5)
plot.new()
plot.window(xlim,ylim)
axis(1,at=pretty(xlim,10),labels=as.character(pretty(xlim,10)))
axis(2,at=pretty(ylim,10))
title(paste(main="Volcano plot",con),
xlab="Fold change",ylab="-log10(p-value)")
points(ff,-log10(pp),pch=20,col="blue2",cex=0.9)
points(f[down],-log10(p[down]),pch=20,col="green3",cex=0.9)
points(f[up],-log10(p[up]),pch=20,col="red2",cex=0.9)
abline(h=-log10(pcut),lty=4)
abline(v=-fcut,lty=2)
abline(v=fcut,lty=2)
grid()
graphics::legend(
x="topleft",
legend=c("up-regulated","down-regulated","unregulated",
"p-value threshold","fold change threshold"),
col=c("red2","green3","blue1","black","black"),
pch=c(20,20,20,NA,NA),lty=c(NA,NA,NA,4,2),
xjust=1,yjust=0,box.lty=0,x.intersp=0.5,cex=0.8,text.font=2
)
graphics.close(output)
}
else {
obj <- list(
x=f,
y=p,
plot=NULL,
samples=NULL,
xlim=xlim,
ylim=ylim,
status=NULL,
pcut=pcut,
fcut=fcut,
altnames=alt.names,
user=list(up=up,down=down,unf=ff,unp=pp,ualt=alt.names.neutral,
con=con)
)
#json <- volcanoToJSON(obj)
#fil <- file.path(path,paste("volcano_",con,".json",sep=""))
#write(json,fil)
fil <- volcanoToJSON(obj)
}
return(fil)
}
#' Diagnostic heatmap of differentially expressed genes
#'
#' This function plots a heatmap of the differentially expressed genes produced
#' by the metaseqr workflow, useful for quality control, e.g. whether samples
#' belonging to the same group cluster together.
#'
#' @param x the data matrix to create a heatmap for.
#' @param con an optional string depicting a name (e.g. the contrast name) to
#' appear in the title of the volcano plot.
#' @param output one or more R plotting device to direct the plot result to.
#' Supported mechanisms: \code{"x11"} (default), \code{"png"}, \code{"jpg"},
#' \code{"bmp"}, \code{"pdf"}, \code{"ps"}.
#' @param path the path to create output files.
#' @param ... further arguments to be passed to plot devices, such as parameter
#' from \code{\link{par}}.
#' @return The filenames of the plots produced in a named list with names the
#' \code{which.plot} argument. If \code{output="x11"}, no output filenames are
#' produced.
#' @export
#' @author Panagiotis Moulos
#' @examples
#' \dontrun{
#' require(DESeq)
#' data.matrix <- counts(makeExampleCountDataSet())
#' sample.list <- list(A=c("A1","A2"),B=c("B1","B2","B3"))
#' contrast <- "A_vs_B"
#' M <- norm.edger(data.matrix,sample.list)
#' p <- stat.edger(M,sample.list,contrast)
#' diagplot.de.heatmap(data.matrix[p[[1]]<0.05])
#'}
diagplot.de.heatmap <- function(x,con=NULL,output="x11",path=NULL,...) {
if (is.null(path)) path <- getwd()
if (is.null(con))
con <- conn <- ""
else {
conn <- con
con <- paste("for ",con)
}
y <- nat2log(x,2,1)
# First plot the normal image
fil <- file.path(path,paste("de_heatmap_",conn,".",output,sep=""))
if (output %in% c("pdf","ps","x11"))
graphics.open(output,fil,width=10,height=10)
else
graphics.open(output,fil,width=800,height=800)
heatmap.2(y,trace="none",col=bluered(16),labRow="",cexCol=0.9,keysize=1,
font.lab=2,main=paste("DEG heatmap",con),cex.main=0.9)
graphics.close(output)
## Then the "interactive" using sendplot
#xy.labels <- list(normalized_counts=x,log2_normalized_counts=y)
#x.labels <- data.frame(
# label=colnames(x),
# description=paste("Sample",colnames(x))
#)
#y.labels <- data.frame(
# label=rownames(x),
# description=paste("Gene ID:",rownames(x))
#)
#suppressWarnings(heatmap.send(
# y,
# distfun=dist,
# hclustfun=hclust,
# MainColor=bluered(16),
# labRow="",
# labCol=NULL,
# keep.dendro=TRUE,
# x.labels=x.labels,
# y.labels=y.labels,
# xy.labels=xy.labels,
# image.size="2048x4096",
# fname.root=paste("iframe_de_heatmap_",conn,sep=""),
# dir=paste(path,.Platform$file.sep,sep=""),
# header="v3",
# window.size="2048x4192"
#))
return(fil)
}
#' Diagnostic plot for filtered genes
#'
#' This function plots a grid of four graphs depicting: in the first row, the
#' numbers of filtered genes per chromosome in the first column and per biotype
#' in the second column. In the second row, the percentages of filtered genes
#' per chromosome related to the whole genome in the first columns and per biotype
#' in the second column.
#'
#' @param x an annotation data frame like the ones produced by
#' \code{\link{get.annotation}}. \code{x} should be the filtered annotation
#' according to metaseqR's filters.
#' @param y an annotation data frame like the ones produced by
#' \code{\link{get.annotation}}. \code{y} should contain the total annotation
#' without the application of any metaseqr filter.
#' @param output one or more R plotting device to direct the plot result to.
#' Supported mechanisms: \code{"x11"} (default), \code{"png"}, \code{"jpg"},
#' \code{"bmp"}, \code{"pdf"} or \code{"ps"}.
#' @param path the path to create output files.
#' @param ... further arguments to be passed to plot devices, such as parameter
#' from \code{\link{par}}.
#' @return The filenames of the plots produced in a named list with names the
#' \code{which.plot} argument. If output=\code{"x11"}, no output filenames are
#' produced.
#' @export
#' @author Panagiotis Moulos
#' @examples
#' \dontrun{
#' y <- get.annotation("mm9","gene")
#' x <- y[-sample(1:nrow(y),10000),]
#' diagplot.filtered(x,y)
#'}
diagplot.filtered <- function(x,y,output="x11",path=NULL,...) {
if (output !="json") {
if (is.null(path)) path <- getwd()
fil <- file.path(path,paste("filtered_genes.",output,sep=""))
if (output %in% c("pdf","ps","x11"))
graphics.open(output,fil,width=12,height=8)
else
graphics.open(output,fil,width=1200,height=800,res=100)
chr <- table(as.character(x$chromosome))
bt <- table(as.character(x$biotype))
chr.all <- table(as.character(y$chromosome))
bt.all <- table(as.character(y$biotype))
barlab.chr <- as.character(chr)
barlab.bt <- as.character(bt)
per.chr <- chr/chr.all[names(chr)]
per.bt <- bt/bt.all[names(bt)]
# Some bug...
per.chr[per.chr>1] <- 1
per.bt[per.bt>1] <- 1
#
suppressWarnings(per.chr.lab <- paste(formatC(100*per.chr,digits=1,
format="f"),"%",sep=""))
suppressWarnings(per.bt.lab <- paste(formatC(100*per.bt,digits=1,
format="f"),"%",sep=""))
par(mfrow=c(2,2),mar=c(1,4,2,1),oma=c(1,1,1,1))
# Chromosomes
barx.chr <- barplot(chr,space=0.5,
ylim=c(0,max(chr)+ceiling(max(chr)/10)),yaxt="n",xaxt="n",
plot=FALSE)
plot.new()
plot.window(xlim=c(0,ceiling(max(barx.chr))),
ylim=c(0,max(chr)+ceiling(max(chr)/10)),mar=c(1,4,1,1))
axis(2,at=pretty(0:(max(chr)+ceiling(max(chr)/10))),cex.axis=0.9,padj=1,
font=2)
text(x=barx.chr,y=chr,label=barlab.chr,cex=0.7,font=2,col="green3",
adj=c(0.5,-1.3))
title(main="Filtered genes per chromosome",cex.main=1.1)
mtext(side=2,text="Number of genes",line=2,cex=0.9,font=2)
grid()
barplot(chr,space=0.5,ylim=c(0,max(chr)+ceiling(max(chr)/10)),
col="blue3",border="yellow3",yaxt="n",xaxt="n",font=2,add=TRUE)
# Biotypes
barx.bt <- barplot(bt,space=0.5,ylim=c(0,max(bt)+ceiling(max(bt)/10)),
yaxt="n",xaxt="n",plot=FALSE)
plot.new()
plot.window(xlim=c(0,ceiling(max(barx.bt))),
ylim=c(0,max(bt)+ceiling(max(bt)/10)),mar=c(1,4,1,1))
axis(2,at=pretty(0:(max(bt)+ceiling(max(bt)/10))),cex.axis=0.9,padj=1,
font=2)
text(x=barx.bt,y=bt,label=barlab.bt,cex=0.7,font=2,col="blue",
adj=c(0.5,-1.3),xpd=TRUE)
title(main="Filtered genes per biotype",cex.main=1.1)
mtext(side=2,text="Number of genes",line=2,cex=0.9,font=2)
grid()
barplot(bt,space=0.5,ylim=c(0,max(bt)+ceiling(max(bt)/10)),col="red3",
border="yellow3",yaxt="n",xaxt="n",font=2,add=TRUE)
# Chromosome percentage
barx.per.chr <- barplot(per.chr,space=0.5,ylim=c(0,max(per.chr)),
yaxt="n",xaxt="n",plot=FALSE)
plot.new()
par(mar=c(9,4,1,1))
plot.window(xlim=c(0,max(barx.per.chr)),ylim=c(0,max(per.chr)))
#axis(1,at=barx.per.chr,labels=names(per.chr),cex.axis=0.9,font=2,
# tcl=-0.3,col="lightgrey",las=2)
axis(1,at=barx.per.chr,labels=FALSE,tcl=-0.3,col="lightgrey")
axis(2,at=seq(0,max(per.chr),length.out=5),labels=formatC(seq(0,
max(per.chr),length.out=5),digits=2,format="f"),cex.axis=0.9,padj=1,
font=2)
text(barx.per.chr,par("usr")[3]-max(per.chr)/17,labels=names(per.chr),
srt=45,adj=c(1,1.1),xpd=TRUE,cex=0.9,font=2)
text(x=barx.per.chr,y=per.chr,label=per.chr.lab,cex=0.7,font=2,
col="green3",adj=c(0.5,-1.3),xpd=TRUE)
mtext(side=2,text="fraction of total genes",line=2,cex=0.9,font=2)
grid()
barplot(per.chr,space=0.5,ylim=c(0,max(per.chr)),col="blue3",
border="yellow3",yaxt="n",xaxt="n",font=2,add=TRUE)
# Biotype percentage
barx.per.bt <- barplot(per.bt,space=0.5,ylim=c(0,max(per.bt)),yaxt="n",
xaxt="n",plot=FALSE)
plot.new()
par(mar=c(9,4,1,1))
plot.window(xlim=c(0,max(barx.per.bt)),ylim=c(0,max(per.bt)))
#axis(1,at=barx.per.bt,labels=names(per.bt),cex.axis=0.9,font=2,
# tcl=-0.3,col="lightgrey",las=2)
axis(1,at=barx.per.bt,labels=FALSE,tcl=-0.3,col="lightgrey")
axis(2,at=seq(0,max(per.bt),length.out=5),
labels=formatC(seq(0,max(per.bt),length.out=5),digits=2,format="f"),
cex.axis=0.9,padj=1,font=2)
text(barx.per.bt,par("usr")[3]-max(per.bt)/17,
labels=gsub("prime","'",names(per.bt)),srt=45,adj=c(1,1.1),
xpd=TRUE,cex=0.9,font=2)
text(x=barx.per.bt,y=per.bt,label=per.bt.lab,cex=0.7,font=2,col="blue",
adj=c(0.5,-1.3),xpd=TRUE)
mtext(side=2,text="fraction of total genes",line=2,cex=0.9,font=2)
grid()
barplot(per.bt,space=0.5,ylim=c(0,max(per.bt)),col="red3",
border="yellow3",yaxt="n",xaxt="n",font=2,add=TRUE)
graphics.close(output)
}
else {
obj <- list(
x=NULL,
y=NULL,
plot=NULL,
samples=NULL,
xlim=NULL,
ylim=NULL,
status=NULL,
pcut=NULL,
fcut=NULL,
altnames=NULL,
user=list(filtered=x,total=y)
)
fil <- list(chromosome=NULL,biotype=NULL)
json <- filteredToJSON(obj,by="chromosome")
fil$chromosome <- file.path(path,"filtered_genes_chromosome.json")
write(json,fil$chromosome)
json <- filteredToJSON(obj,by="biotype")
fil$biotype <- file.path(path,"filtered_genes_biotype.json")
write(json,fil$biotype)
}
return(fil)
}
#' Venn diagrams when performing meta-analysis
#'
#' This function uses the R package VennDiagram and plots an up to 5-way Venn
#' diagram depicting the common and specific to each statistical algorithm genes,
#' for each contrast. Mostly for internal use because of its main argument which
#' is difficult to construct, but can be used independently if the user grasps
#' the logic.
#'
#' @param pmat a matrix with p-values corresponding to the application of each
#' statistical algorithm. The p-value matrix must have the colnames attribute and
#' the colnames should correspond to the name of the algorithm used to fill the
#' specific column (e.g. if \code{"statistics"=c("deseq","edger","nbpseq")} then
#' \code{colnames(pmat) <-} \code{c("deseq","edger","nbpseq")}.
#' @param fcmat an optional matrix with fold changes corresponding to the application
#' of each statistical algorithm. The fold change matrix must have the colnames
#' attribute and the colnames should correspond to the name of the algorithm used
#' to fill the specific column (see the parameter \code{pmat}).
#' @param pcut a p-value cutoff for statistical significance. Defaults to
#' \code{0.05}.
#' @param fcut if \code{fcmat} is supplied, an absolute fold change cutoff to be
#' applied to \code{fcmat} to determine the differentially expressed genes for
#' each algorithm.
#' @param direction if \code{fcmat} is supplied, a keyword to denote which genes
#' to draw in the Venn diagrams with respect to their direction of regulation. It
#' can be one of \code{"dereg"} for the total of regulated genes, where
#' \code{abs(fcmat[,n])>=fcut} (default), \code{"up"} for the up-regulated genes
#' where \code{fcmat[,n]>=fcut} or \code{"down"} for the up-regulated genes where
#' \code{fcmat[,n]<=-fcut}.
#' @param nam a name to be appended to the output graphics file (if \code{"output"}
#' is not \code{"x11"}).
#' @param output one or more R plotting device to direct the plot result to.
#' Supported mechanisms: \code{"x11"} (default), \code{"png"}, \code{"jpg"},
#' \code{"bmp"}, \code{"pdf"} or \code{"ps"}.
#' @param path the path to create output files.
#' @param alt.names an optional named vector of names, e.g. HUGO gene symbols,
#' alternative or complementary to the unique gene names which are the rownames
#' of \code{pmat}. The names of the vector must be the rownames of \code{pmat}.
#' @param ... further arguments to be passed to plot devices, such as parameter
#' from \code{\link{par}}.
#' @return The filenames of the plots produced in a named list with names the
#' \code{which.plot} argument. If output=\code{"x11"}, no output filenames are
#' produced. If \code{"path"} is not \code{NULL}, a file with the intersections
#' in the Venn diagrams will be produced and written in \code{"path"}.
#' @export
#' @author Panagiotis Moulos
#' @examples
#' \dontrun{
#' p1 <- 0.001*matrix(runif(300),100,3)
#' p2 <- matrix(runif(300),100,3)
#' p <- rbind(p1,p2)
#' rownames(p) <- paste("gene",1:200,sep="_")
#' colnames(p) <- paste("method",1:3,sep="_")
#' venn.contents <- diagplot.venn(p)
#'}
diagplot.venn <- function(pmat,fcmat=NULL,pcut=0.05,fcut=0.5,
direction=c("dereg","up","down"),nam=as.character(round(1000*runif(1))),
output="x11",path=NULL,alt.names=NULL,...) {
check.text.args("direction",direction,c("dereg","up","down"))
if (is.na(pcut) || is.null(pcut) || pcut==1)
warnwrap("Illegal pcut argument! Using the default (0.05)")
algs <- colnames(pmat)
if (is.null(algs))
stopwrap("The p-value matrices must have the colnames attribute ",
"(names of statistical algorithms)!")
if (!is.null(fcmat) && (is.null(colnames(fcmat)) ||
length(intersect(colnames(pmat),colnames(fcmat)))!=length(algs)))
stopwrap("The fold change matrices must have the colnames attribute ",
"(names of statistical algorithms) and must be the same as in the ",
"p-value matrices!")
nalg <- length(algs)
if(nalg>5) {
warnwrap(paste("Cannot create a Venn diagram for more than 5 result ",
"sets! ",nalg,"found, only the first 5 will be used..."))
algs <- algs[1:5]
nalg <- 5
}
lenalias <- c("two","three","four","five")
aliases <- toupper(letters[1:nalg])
names(algs) <- aliases
genes <- rownames(pmat)
pairs <- make.venn.pairs(algs)
areas <- make.venn.areas(length(algs))
counts <- make.venn.counts(length(algs))
# Initially populate the results and counts lists so they can be used to create
# the rest of the intersections
results <- vector("list",nalg+length(pairs))
names(results)[1:nalg] <- aliases
names(results)[(nalg+1):length(results)] <- names(pairs)
if (is.null(fcmat)) {
for (a in aliases) {
results[[a]] <- genes[which(pmat[,algs[a]]<pcut)]
counts[[areas[[a]]]] <- length(results[[a]])
}
}
else {
switch(direction,
dereg = {
for (a in aliases) {
results[[a]] <-
genes[which(pmat[,algs[a]]<pcut & abs(
fcmat[,algs[a]])>=fcut)]
counts[[areas[[a]]]] <- length(results[[a]])
}
},
up = {
for (a in aliases) {
results[[a]] <-
genes[which(pmat[,algs[a]]<pcut &
fcmat[,algs[a]]>=fcut)]
counts[[areas[[a]]]] <- length(results[[a]])
}
},
down = {
for (a in aliases) {
results[[a]] <-
genes[which(pmat[,algs[a]]<pcut &
fcmat[,algs[a]]<=-fcut)]
counts[[areas[[a]]]] <- length(results[[a]])
}
}
)
}
# Now, perform the intersections
for (p in names(pairs)) {
a = pairs[[p]][1]
b = pairs[[p]][2]
results[[p]] <- intersect(results[[a]],results[[b]])
counts[[areas[[p]]]] <- length(results[[p]])
}
# And now, the Venn diagrams must be constructed
color.scheme <- make.venn.colorscheme(length(algs))
fil <- file.path(path,paste("venn_plot_",nam,".",output,sep=""))
if (output %in% c("pdf","ps","x11"))
graphics.open(output,fil,width=8,height=8)
else
graphics.open(output,fil,width=800,height=800,res=100)
switch(lenalias[length(algs)-1],
two = {
v <- draw.pairwise.venn(
area1=counts$area1,
area2=counts$area2,
cross.area=counts$cross.area,
category=paste(algs," (",aliases,")",sep=""),
lty="blank",
fill=color.scheme$fill,
cex=1.5,
cat.cex=1.3,
cat.pos=c(0,0),
cat.col=color.scheme$font,
#cat.dist=0.07,
cat.fontfamily=rep("Bookman",2)
)
},
three = {
#overrideTriple <<- TRUE
v <- draw.triple.venn(
area1=counts$area1,
area2=counts$area2,
area3=counts$area3,
n12=counts$n12,
n13=counts$n13,
n23=counts$n23,
n123=counts$n123,
category=paste(algs," (",aliases,")",sep=""),
lty="blank",
fill=color.scheme$fill,
cex=1.5,
cat.cex=1.3,
#cat.pos=c(0,0,180),
cat.col=color.scheme$font,
#cat.dist=0.07,
cat.fontfamily=rep("Bookman",3)
)
},
four = {
v <- draw.quad.venn(
area1=counts$area1,
area2=counts$area2,
area3=counts$area3,
area4=counts$area4,
n12=counts$n12,
n13=counts$n13,
n14=counts$n14,
n23=counts$n23,
n24=counts$n24,
n34=counts$n34,
n123=counts$n123,
n124=counts$n124,
n134=counts$n134,
n234=counts$n234,
n1234=counts$n1234,
category=paste(algs," (",aliases,")",sep=""),
lty="blank",
fill=color.scheme$fill,
cex=1.5,
cat.cex=1.3,
c(0.1,0.1,0.05,0.05),
cat.col=color.scheme$font,
cat.fontfamily=rep("Bookman",4)
)
},
five = {
v <- draw.quintuple.venn(
area1=counts$area1,
area2=counts$area2,
area3=counts$area3,
area4=counts$area4,
area5=counts$area5,
n12=counts$n12,
n13=counts$n13,
n14=counts$n14,
n15=counts$n15,
n23=counts$n23,
n24=counts$n24,
n25=counts$n25,
n34=counts$n34,
n35=counts$n35,
n45=counts$n45,
n123=counts$n123,
n124=counts$n124,
n125=counts$n125,
n134=counts$n134,
n135=counts$n135,
n145=counts$n145,
n234=counts$n234,
n235=counts$n235,
n245=counts$n245,
n345=counts$n345,
n1234=counts$n1234,
n1235=counts$n1235,
n1245=counts$n1245,
n1345=counts$n1345,
n2345=counts$n2345,
n12345=counts$n12345,
category=paste(algs," (",aliases,")",sep=""),
lty="blank",
fill=color.scheme$fill,
cex=1.5,
cat.cex=1.3,
cat.dist=0.1,
cat.col=color.scheme$font,
cat.fontfamily=rep("Bookman",5)
)
}
)
graphics.close(output)
# Now do something with the results
if (!is.null(path)) {
results.ex <- vector("list",length(results))
names(results.ex) <- names(results)
if (!is.null(alt.names)) {
for (n in names(results))
results.ex[[n]] <- alt.names[results[[n]]]
}
else {
for (n in names(results))
results.ex[[n]] <- results[[n]]
}
max.len <- max(sapply(results.ex,length))
for (n in names(results.ex)) {
if (length(results.ex[[n]])<max.len) {
dif <- max.len - length(results.ex[[n]])
results.ex[[n]] <- c(results.ex[[n]],rep(NA,dif))
}
}
results.ex <- do.call("cbind",results.ex)
write.table(results.ex,file=file.path(path,"..","..","lists",
paste0("venn_categories_",nam,".txt")),sep="\t",
row.names=FALSE,quote=FALSE,na="")
}
return(fil)
}
#' Helper for Venn diagrams
#'
#' This function creates a list of pairwise comparisons to be performed in order
#' to create an up to 5-way Venn diagram using the R package VennDiagram. Internal
#' use mostly.
#'
#' @param algs a vector with the names of the sets (up to length 5, if larger, it
#' will be truncated with a warning).
#' @return A list with as many pairs as the comparisons to be made for the
#' construction of the Venn diagram. The pairs are encoded with the uppercase
#' letters A through E, each one corresponding to order of the input sets.
#' @export
#' @author Panagiotis Moulos
#' @examples
#' \dontrun{
#' sets <- c("apple","pear","banana")
#' pairs <- make.venn.pairs(sets)
#'}
make.venn.pairs <- function(algs)
{
lenalias <- c("two","three","four","five")
switch(lenalias[length(algs)-1],
two = {
return(list(
AB=c("A","B")
))
},
three = {
return(list(
AB=c("A","B"),
AC=c("A","C"),
BC=c("B","C"),
ABC=c("AB","C")
))
},
four = {
return(list(
AB=c("A","B"),
AC=c("A","C"),
AD=c("A","D"),
BC=c("B","C"),
BD=c("B","D"),
CD=c("C","D"),
ABC=c("AB","C"),
ABD=c("AB","D"),
ACD=c("AC","D"),
BCD=c("BC","D"),
ABCD=c("ABC","D")
))
},
five = {
return(list(
AB=c("A","B"),
AC=c("A","C"),
AD=c("A","D"),
AE=c("A","E"),
BC=c("B","C"),
BD=c("B","D"),
BE=c("B","E"),
CD=c("C","D"),
CE=c("C","E"),
DE=c("D","E"),
ABC=c("AB","C"),
ABD=c("AB","D"),
ABE=c("AB","E"),
ACD=c("AC","D"),
ACE=c("AC","E"),
ADE=c("AD","E"),
BCD=c("BC","D"),
BCE=c("BC","E"),
BDE=c("BD","E"),
CDE=c("CD","E"),
ABCD=c("ABC","D"),
ABCE=c("ABC","E"),
ABDE=c("ABD","E"),
ACDE=c("ACD","E"),
BCDE=c("BCD","E"),
ABCDE=c("ABCD","E")
))
}
)
}
#' Helper for Venn diagrams
#'
#' This function creates a list with names the arguments of the Venn diagram
#' construction functions of the R package VennDiagram and list members the
#' internal encoding (uppercase letters A to E and combinations among then) used
#' to encode the pairwise comparisons to create the intersections needed for the
#' Venn diagrams. Internal use mostly.
#'
#' @param n the number of the sets used for the Venn diagram.
#' @return A named list, see descritpion.
#' @export
#' @author Panagiotis Moulos
#' @examples
#' \dontrun{
#' sets <- c("apple","pear","banana")
#' pairs <- make.venn.pairs(sets)
#' areas <- make.venn.areas(length(sets))
#'}
make.venn.areas <- function(n)
{
lenalias <- c("two","three","four","five")
switch(lenalias[n-1],
two = {
return(list(
A="area1",
B="area2",
AB="cross.area"
))
},
three = {
return(list(
A="area1",
B="area2",
C="area3",
AB="n12",
AC="n13",
BC="n23",
ABC="n123"
))
},
four = {
return(list(
A="area1",
B="area2",
C="area3",
D="area4",
AB="n12",
AC="n13",
AD="n14",
BC="n23",
BD="n24",
CD="n34",
ABC="n123",
ABD="n124",
ACD="n134",
BCD="n234",
ABCD="n1234"
))
},
five = {
return(list(
A="area1",
B="area2",
C="area3",
D="area4",
E="area5",
AB="n12",
AC="n13",
AD="n14",
AE="n15",
BC="n23",
BD="n24",
BE="n25",
CD="n34",
CE="n35",
DE="n45",
ABC="n123",
ABD="n124",
ABE="n125",
ACD="n134",
ACE="n135",
ADE="n145",
BCD="n234",
BCE="n235",
BDE="n245",
CDE="n345",
ABCD="n1234",
ABCE="n1235",
ABDE="n1245",
ACDE="n1345",
BCDE="n2345",
ABCDE="n12345"
))
}
)
}
#' Helper for Venn diagrams
#'
#' This function creates a list with names the arguments of the Venn diagram
#' construction functions of the R package VennDiagram and list members are
#' initially \code{NULL}. They are filled by the \code{\link{diagplot.venn}}
#' function. Internal use mostly.
#'
#' @param n the number of the sets used for the Venn diagram.
#' @return A named list, see descritpion.
#' @export
#' @author Panagiotis Moulos
#' @examples
#' \dontrun{
#' sets <- c("apple","pear","banana")
#' counts <- make.venn.counts(length(sets))
#'}
make.venn.counts <- function(n)
{
lenalias <- c("two","three","four","five")
switch(lenalias[n-1],
two = {
return(list(
area1=NULL,
area2=NULL,
cross.area=NULL
))
},
three = {
return(list(
area1=NULL,
area2=NULL,
area3=NULL,
n12=NULL,
n13=NULL,
n23=NULL,
n123=NULL
))
},
four = {
return(list(
area1=NULL,
area2=NULL,
area3=NULL,
area4=NULL,
n12=NULL,
n13=NULL,
n14=NULL,
n23=NULL,
n24=NULL,
n34=NULL,
n123=NULL,
n124=NULL,
n134=NULL,
n234=NULL,
n1234=NULL
))
},
five = {
return(list(
area1=NULL,
area2=NULL,
area3=NULL,
area4=NULL,
area5=NULL,
n12=NULL,
n13=NULL,
n14=NULL,
n15=NULL,
n23=NULL,
n24=NULL,
n25=NULL,
n34=NULL,
n35=NULL,
n45=NULL,
n123=NULL,
n124=NULL,
n125=NULL,
n134=NULL,
n135=NULL,
n145=NULL,
n234=NULL,
n235=NULL,
n245=NULL,
n345=NULL,
n1234=NULL,
n1235=NULL,
n1245=NULL,
n1345=NULL,
n2345=NULL,
n12345=NULL
))
}
)
}
#' Helper for Venn diagrams
#'
#' This function returns a list of colorschemes accroding to the number of sets.
#' Internal use.
#'
#' @param n the number of the sets used for the Venn diagram.
#' @return A list with colors for fill and font.
#' @export
#' @author Panagiotis Moulos
#' @examples
#' \dontrun{
#' sets <- c("apple","pear","banana")
#' cs <- make.venn.colorscheme(length(sets))
#'}
make.venn.colorscheme <- function(n) {
lenalias <- c("two","three","four","five")
switch(lenalias[n-1],
two = {
return(list(
fill=c("blue","orange2"),
font=c("darkblue","orange4")
))
},
three = {
return(list(
fill=c("red","green","mediumpurple"),
font=c("darkred","darkgreen","mediumpurple4")
))
},
four = {
return(list(
fill=c("red","green","mediumpurple","orange2"),
font=c("darkred","darkgreen","mediumpurple4","orange4")
))
},
five = {
return(list(
fill=c("red","green","blue","mediumpurple","orange2"),
font=c("darkred","darkgreen","darkblue","mediumpurple4",
"orange4")
))
}
)
}
#' Create basic ROC curves
#'
#' This function creates basic ROC curves using a matrix of p-values (such a matrix
#' can be derived for example from the result table of \code{\link{metaseqr}} by
#' subsetting the table to get the p-values from several algorithms) given a ground
#' truth vector for differential expression and a significance level.
#'
#' @param truth the ground truth differential expression vector. It should contain
#' only zero and non-zero elements, with zero denoting non-differentially expressed
#' genes and non-zero, differentially expressed genes. Such a vector can be obtained
#' for example by using the \code{\link{make.sim.data.sd}} function, which creates
#' simulated RNA-Seq read counts based on real data.
#' @param p a p-value matrix whose rows correspond to each element in the
#' \code{truth} vector. If the matrix has a \code{colnames} attribute, a legend
#' will be added to the plot using these names, else a set of column names will
#' be auto-generated. \code{p} can also be a list or a data frame.
#' @param sig a significance level (0 < \code{sig} <=1).
#' @param x what to plot on x-axis, can be one of \code{"fpr"}, \code{"fnr"},
#' \code{"tpr"}, \code{"tnr"} for False Positive Rate, False Negative Rate, True
#' Positive Rate and True Negative Rate respectively.
#' @param y what to plot on y-axis, same as \code{x} above.
#' @param output one or more R plotting device to direct the plot result to.
#' Supported mechanisms: \code{"x11"} (default), \code{"png"}, \code{"jpg"},
#' \code{"bmp"}, \code{"pdf"} or \code{"ps"}.
#' @param path the path to create output files.
#' @param draw boolean to determine whether to plot the curves or just return the
#' calculated values (in cases where the user wants the output for later averaging
#' for example). Defaults to \code{TRUE} (make plots).
#' @param ... further arguments to be passed to plot devices, such as parameter
#' from \code{\link{par}}.
#' @return A named list with two members. The first member is a list containing
#' the ROC statistics: \code{TP} (True Postives), \code{FP} (False Positives),
#' \code{FN} (False Negatives), \code{TN} (True Negatives), \code{FPR} (False
#' Positive Rate), \code{FNR} (False Negative Rate), \code{TPR} (True Positive
#' Rate), \code{TNR} (True Negative Rate), \code{AUC} (Area Under the Curve). The
#' second is the path to the created figure graphic.
#' @export
#' @author Panagiotis Moulos
#' @examples
#' \dontrun{
#' p1 <- 0.001*matrix(runif(300),100,3)
#' p2 <- matrix(runif(300),100,3)
#' p <- rbind(p1,p2)
#' rownames(p) <- paste("gene",1:200,sep="_")
#' colnames(p) <- paste("method",1:3,sep="_")
#' truth <- c(rep(1,40),rep(-1,40),rep(0,20),rep(1,10),rep(2,10),rep(0,80))
#' names(truth) <- rownames(p)
#' roc.obj <- diagplot.roc(truth,p)
#'}
diagplot.roc <- function(truth,p,sig=0.05,x="fpr",y="tpr",output="x11",
path=NULL,draw=TRUE,...) {
check.text.args("x",x,c("fpr","fnr","tpr","tnr","scrx","sens","spec"),
multiarg=FALSE)
check.text.args("y",y,c("fpr","fnr","tpr","tnr","scry","sens","spec"),
multiarg=FALSE)
if (is.list(p))
pmat <- do.call("cbind",p)
else if (is.data.frame(p))
pmat <- as.matrix(p)
else if (is.matrix(p))
pmat <- p
if (is.null(colnames(pmat)))
colnames(pmat) <- paste("p",1:ncol(pmat),sep="_")
ax.name <- list(
tpr="True Positive Rate",
tnr="True Negative Rate",
fpr="False Positive Rate",
fnr="False Negative Rate",
scrx="Ratio of selected",
scry="Normalized TP/(FP+FN)",
sens="Sensitivity",
spec="1 - Specificity"
)
ROC <- vector("list",ncol(pmat))
names(ROC) <- colnames(pmat)
colspace.universe <- c("red","blue","green","orange","darkgrey","green4",
"black","pink","brown","magenta","yellowgreen","pink4","seagreen4",
"darkcyan")
colspace <- colspace.universe[1:ncol(pmat)]
names(colspace) <- colnames(pmat)
eps <- min(pmat[!is.na(pmat) & pmat>0])
for (n in colnames(pmat)) {
disp("Processing ",n)
gg <- which(pmat[,n]<=sig)
psample <- -log10(pmax(pmat[gg,n],eps))
#psample <- pmat[gg,n]
size <- seq(1,length(gg))
cuts <- seq(-log10(sig),max(psample),length.out=length(gg))
#cuts <- seq(min(psample),sig,length.out=length(gg))
local.truth <- truth[gg]
S <- length(size)
TP <- FP <- FN <- TN <- FPR <- FNR <- TPR <- TNR <- SENS <- SPEC <-
SCRX <- SCRY <- numeric(S)
for (i in 1:S) {
TP[i] <- length(which(psample>cuts[i] & local.truth!=0))
FP[i] <- length(which(psample>cuts[i] & local.truth==0))
FN[i] <- length(which(psample<cuts[i] & local.truth!=0))
TN[i] <- length(which(psample<cuts[i] & local.truth==0))
## Alternatives which I keep in the code
#TP[i] <- length(intersect(names(which(psample>cuts[i])),
# names(which(local.truth!=0))))
#FP[i] <- length(intersect(names(which(psample>cuts[i])),
# names(which(local.truth==0))))
#FN[i] <- length(intersect(names(which(psample<cuts[i])),
# names(which(local.truth!=0))))
#TN[i] <- length(intersect(names(which(psample<cuts[i])),
# names(which(local.truth==0))))
#bad <- which(psample<cuts[i])
#good <- which(psample>cuts[i])
#TP[i] <- length(which(local.truth[good]!=0))
#FP[i] <- length(which(local.truth[good]==0))
#TN[i] <- length(which(local.truth[bad]==0))
#FN[i] <- length(which(local.truth[bad]!=0))
SCRX[i] <- i/S
SCRY[i] <- TP[i]/(FN[i]+FP[i])
if (FP[i]+TN[i] == 0)
FPR[i] <- 0
else
FPR[i] <- FP[i]/(FP[i]+TN[i])
FNR[i] <- FN[i]/(TP[i]+FN[i])
TPR[i] <- TP[i]/(TP[i]+FN[i])
if (TN[i]+FP[i] == 0)
TNR[i] <- 0
else
TNR[i] <- TN[i]/(TN[i]+FP[i])
SENS[i] <- TPR[i]
SPEC[i] <- 1 - TNR[i]
}
#if (all(FPR==0))
# FPR[length(FPR)] <- 1
#if (all(TNR==0)) {
# TNR[1] <- 1
# SPEC[i] <- 0
#}
ROC[[n]] <- list(TP=TP,FP=FP,FN=FN,TN=TN,
FPR=FPR,FNR=FNR,TPR=TPR,TNR=TNR,SCRX=SCRX,SCRY=SCRY/max(SCRY),
SENS=SENS,SPEC=SPEC,AUC=NULL)
}
for (n in colnames(pmat)) {
disp("Calculating AUC for ",n)
auc <- 0
for (i in 2:length(ROC[[n]][[toupper(y)]])) {
auc <- auc +
0.5*(ROC[[n]][[toupper(x)]][i]-ROC[[n]][[toupper(x)]][i-1])*
(ROC[[n]][[toupper(y)]][i]+ROC[[n]][[toupper(y)]][i-1])
}
ROC[[n]]$AUC <- abs(auc)
# There are some extreme cases, with the Intersection case for the paper
# where there are no FPs or TNs for a p-value cutoff of 0.2 (which is
# imposed in order to avoid the saturation of the ROC curves). In these
# cases, performance is virtually perfect, and the actual AUC should be
# 1. For these cases, we set it to a value between 0.95 and 0.99 to
# represent a more plausible truth.
if (ROC[[n]]$AUC==0) ROC[[n]]$AUC <- sample(seq(0.95,0.99,by=0.001),1)
}
disp("")
if (draw) {
fil <- file.path(path,paste("ROC",output,sep="."))
if (output %in% c("pdf","ps","x11"))
graphics.open(output,fil,width=8,height=8)
else
graphics.open(output,fil,width=1024,height=1024,res=100)
xlim <- c(0,1)
ylim <- c(0,1)
par(cex.axis=0.9,cex.main=1,cex.lab=0.9,font.lab=2,font.axis=2,pty="m",
lwd=1.5,lty=1)
plot.new()
plot.window(xlim,ylim)
axis(1,at=pretty(xlim,10))
axis(2,at=pretty(ylim,10))
for (n in names(ROC))
lines(ROC[[n]][[toupper(x)]],ROC[[n]][[toupper(y)]],
col=colspace[n],...)
grid()
title(xlab=ax.name[[x]],ylab=ax.name[[y]])
auc.text <- as.character(sapply(ROC,function(x)
round(x$AUC,digits=3)))
graphics::legend(x="bottomright",col=colspace,lty=1,cex=0.9,
legend=paste(names(ROC)," (AUC = ",auc.text,")",sep=""))
graphics.close(output)
}
else
fil <- NULL
return(list(ROC=ROC,truth=truth,sig.level=sig,x.axis=x,y.axis=y,path=fil))
}
## Create averaged basic ROC curves
##
## This function creates averaged basic ROC curves using a list of objects returned
## from the \code{\link{diagplot.roc}} function.
##
## @param roc.obj a list containing several lists returned from the application
## of \code{\link{diagplot.roc}} function.
## @param output one or more R plotting device to direct the plot result to.
## Supported mechanisms: \code{"x11"} (default), \code{"png"}, \code{"jpg"},
## \code{"bmp"}, \code{"pdf"} or \code{"ps"}.
## @param path the path to create output files.
## @param ... further arguments to be passed to plot devices, such as parameter
## from \code{\link{par}}.
## @return A named list with two members. The first member is a list containing
## the mean and standard deviation of ROC statistics. The second is the path to
## the created figure graphic.
## @export
## @author Panagiotis Moulos
## @examples
## \dontrun{
## # Not yet available
##}
#diagplot.avg.roc <- function(roc.obj,output="x11",path=NULL,...) {
# ax.name <- list(
# tpr="True Positive Rate",
# tnr="True Negative Rate",
# fpr="False Positive Rate",
# fnr="False Negative Rate"
# )
# stats <- names(roc.obj[[1]]$ROC)
# x <- toupper(roc.obj[[1]]$x.axis)
# y <- toupper(roc.obj[[1]]$y.axis)
# avg.ROC <- vector("list",length(stats))
# avg.ROC <- lapply(avg.ROC,function(x) {
# return(list(TP=NULL,FP=NULL,FN=NULL,TN=NULL,
# FPR=NULL,FNR=NULL,TPR=NULL,TNR=NULL,AUC=NULL))
# })
# names(avg.ROC) <- stats
# colspace.universe <- c("red","blue","green","orange","darkgrey","green4",
# "black","pink","brown","yellowgreen","magenta","pink2","seagreen4",
# "darkcyan")
# colspace <- colspace.universe[1:length(stats)]
# names(colspace) <- stats
# for (s in stats) {
# disp("Retrieving ",s)
# for (r in names(avg.ROC[[s]])) {
# if (r != "AUC") {
# #avg.ROC[[s]][[r]] <- do.call("cbind",lapply(roc.obj,
# # function(x,s,r) x$ROC[[s]][[r]],s,r))
# lapply(roc.obj,function(x,s,r) print(length(x$ROC[[s]][[r]])),s,r)
# mn <- apply(avg.ROC[[s]][[r]],1,mean)
# st <- apply(avg.ROC[[s]][[r]],1,sd)
# avg.ROC[[s]][[r]] <- list(mean=mn,sd=st)
# }
# }
# }
# disp("")
# means <- do.call("cbind",lapply(avg.ROC,function(x) x$mean))
# stds <- do.call("cbind",lapply(avg.ROC,function(x) x$sd))
# fil <- file.path(path,paste("ROC",output,sep="."))
# if (output %in% c("pdf","ps","x11"))
# graphics.open(output,fil,width=8,height=8)
# else
# graphics.open(output,fil,width=1024,height=1024,res=100)
# xlim <- c(0,1)
# ylim <- c(0,1)
# par(cex.axis=0.9,cex.main=1,cex.lab=0.9,font.lab=2,font.axis=2,pty="m",
# lwd=1.5,lty=1)
# plot.new()
# plot.window(xlim,ylim)
# axis(1,at=pretty(xlim,10))
# axis(2,at=pretty(ylim,10))
# for (n in names(ROC)) {
# lines(ROC[[n]][[x]],ROC[[n]][[y]],col=colspace[n],...)
# }
# grid()
# title(xlab=ax.name[[x]],ylab=ax.name[[y]])
# legend(x="bottomright",legend=names(ROC),col=colspace,lty=1)
# graphics.close(output)
# return(list(ROC=ROC,path=fil))
#}
#' Create False (or True) Positive (or Negative) curves
#'
#' This function creates false (or true) discovery curves using a matrix of
#' p-values (such a matrix can be derived for example from the result table of
#' \code{\link{metaseqr}} by subsetting the table to get the p-values from several
#' algorithms) given a ground truth vector for differential expression.
#'
#' @param truth the ground truth differential expression vector. It should contain
#' only zero and non-zero elements, with zero denoting non-differentially expressed
#' genes and non-zero, differentially expressed genes. Such a vector can be obtained
#' for example by using the \code{\link{make.sim.data.sd}} function, which creates
#' simulated RNA-Seq read counts based on real data. The elements of \code{truth}
#' MUST be named (e.g. each gene's name).
#' @param p a p-value matrix whose rows correspond to each element in the
#' \code{truth} vector. If the matrix has a \code{colnames} attribute, a legend
#' will be added to the plot using these names, else a set of column names will
#' be auto-generated. \code{p} can also be a list or a data frame. The p-values
#' MUST be named (e.g. each gene's name).
#' @param type what to plot, can be \code{"fpc"} for False Positive Curves
#' (default), \code{"tpc"} for True Positive Curves, \code{"fnc"} for False
#' Negative Curves or \code{"tnc"} for True Negative Curves.
#' @param N create the curves based on the top (or bottom) \code{N} ranked genes
#' (default is 2000) to be used with \code{type="fpc"} or \code{type="tpc"}.
#' @param output one or more R plotting device to direct the plot result to.
#' Supported mechanisms: \code{"x11"} (default), \code{"png"}, \code{"jpg"},
#' \code{"bmp"}, \code{"pdf"} or \code{"ps"}.
#' @param path the path to create output files.
#' @param draw boolean to determine whether to plot the curves or just return the
#' calculated values (in cases where the user wants the output for later averaging
#' for example). Defaults to \code{TRUE} (make plots).
#' @param ... further arguments to be passed to plot devices, such as parameter
#' from \code{\link{par}}.
#' @return A named list with two members: the first member (\code{ftdr}) contains
#' the values used to create the plot. The second member (\code{path}) contains
#' the path to the created figure graphic.
#' @export
#' @author Panagiotis Moulos
#' @examples
#' \dontrun{
#' p1 <- 0.001*matrix(runif(300),100,3)
#' p2 <- matrix(runif(300),100,3)
#' p <- rbind(p1,p2)
#' rownames(p) <- paste("gene",1:200,sep="_")
#' colnames(p) <- paste("method",1:3,sep="_")
#' truth <- c(rep(1,40),rep(-1,40),rep(0,10),rep(1,10),rep(2,10),rep(0,80))
#' names(truth) <- rownames(p)
#' ftd.obj <- diagplot.ftd(truth,p,N=100)
#'}
diagplot.ftd <- function(truth,p,type="fpc",N=2000,output="x11",path=NULL,
draw=TRUE,...) {
check.text.args("type",type,c("fpc","tpc","fnc","tnc"),multiarg=FALSE)
if (is.list(p))
pmat <- do.call("cbind",p)
else if (is.data.frame(p))
pmat <- as.matrix(p)
else if (is.matrix(p))
pmat <- p
else if (is.numeric(p))
pmat <- as.matrix(p)
if (is.null(colnames(pmat)))
colnames(pmat) <- paste("p",1:ncol(pmat),sep="_")
y.name <- list(
tpc="Number of True Positives",
fpc="Number of False Positives",
tnc="Number of True Negatives",
fnc="Number of False Negatives"
)
ftdr.list <- vector("list",ncol(pmat))
names(ftdr.list) <- colnames(pmat)
colspace.universe <- c("red","blue","green","orange","darkgrey","green4",
"black","pink","brown","magenta","yellowgreen","pink4","seagreen4",
"darkcyan")
colspace <- colspace.universe[1:ncol(pmat)]
names(colspace) <- colnames(pmat)
switch(type,
fpc = {
for (n in colnames(pmat)) {
disp("Processing ",n)
z <- sort(pmat[,n])
for (i in 1:N) {
nn <- length(intersect(names(z[1:i]),
names(which(truth==0))))
if (nn==0)
ftdr.list[[n]][i] <- 1
else
ftdr.list[[n]][i] <- nn
}
}
},
tpc = {
for (n in colnames(pmat)) {
disp("Processing ",n)
z <- sort(pmat[,n])
for (i in 1:N)
ftdr.list[[n]][i] <- length(intersect(names(z[1:i]),
names(which(truth!=0))))
}
},
fnc = {
for (n in colnames(pmat)) {
disp("Processing ",n)
z <- sort(pmat[,n],decreasing=TRUE)
for (i in 1:N) {
nn <- length(intersect(names(z[1:i]),
names(which(truth!=0))))
if (nn==0)
ftdr.list[[n]][i] <- 1
else
ftdr.list[[n]][i] <- nn
}
}
},
tnc = {
for (n in colnames(pmat)) {
disp("Processing ",n)
z <- sort(pmat[,n],decreasing=TRUE)
for (i in 1:N)
ftdr.list[[n]][i] <- length(intersect(names(z[1:i]),
names(which(truth==0))))
}
}
)
disp("")
if (draw) {
fil <- file.path(path,paste("FTDR_",type,".",output,sep=""))
if (output %in% c("pdf","ps","x11"))
graphics.open(output,fil,width=8,height=8)
else
graphics.open(output,fil,width=1024,height=1024,res=100)
xlim <- ylim <- c(1,N)
#ylim <- c(1,length(which(truth!=0)))
par(cex.axis=0.9,cex.main=1,cex.lab=0.9,font.lab=2,font.axis=2,pty="m",
lwd=1.5,lty=1)
plot.new()
switch(type,
fpc = {
plot.window(xlim,ylim,log="y")
axis(1,at=pretty(xlim,10))
axis(2)
for (n in names(ftdr.list)) {
lines(ftdr.list[[n]],col=colspace[n],...)
}
grid()
title(main="Selected genes vs False Positives",
xlab="Number of selected genes",ylab=y.name[[type]])
graphics::legend(x="topleft",legend=names(ftdr.list),
col=colspace,lty=1)
},
tpc = {
plot.window(xlim,ylim)
axis(1,at=pretty(xlim,10))
axis(2,at=pretty(ylim,10))
for (n in names(ftdr.list)) {
lines(ftdr.list[[n]],col=colspace[n],...)
}
grid()
title(main="Selected genes vs True Positives",
xlab="Number of selected genes",ylab=y.name[[type]])
graphics::legend(x="bottomright",legend=names(ftdr.list),
col=colspace,lty=1)
},
fnc = {
plot.window(xlim,ylim,log="y")
axis(1,at=pretty(xlim,10))
axis(2)
for (n in names(ftdr.list)) {
lines(ftdr.list[[n]],col=colspace[n],...)
}
grid()
title(main="Selected genes vs False Negatives",
xlab="Number of selected genes",ylab=y.name[[type]])
graphics::legend(x="topleft",legend=names(ftdr.list),
col=colspace,lty=1)
},
tnc = {
plot.window(xlim,ylim)
axis(1,at=pretty(xlim,10))
axis(2,at=pretty(ylim,10))
for (n in names(ftdr.list)) {
lines(ftdr.list[[n]],col=colspace[n],...)
}
grid()
title(main="Selected genes vs True Negatives",
xlab="Number of selected genes",ylab=y.name[[type]])
graphics::legend(x="bottomright",legend=names(ftdr.list),
col=colspace,lty=1)
}
)
graphics.close(output)
}
else
fil <- NULL
return(list(ftdr=ftdr.list,truth=truth,type=type,N=N,path=fil))
}
#' Create average False (or True) Discovery curves
#'
#' This function creates false (or true) discovery curves using a list containing
#' several outputs from \code{\link{diagplot.ftd}}.
#'
#' @param ftdr.obj a list with outputs from \code{\link{diagplot.ftd}}.
#' @param output one or more R plotting device to direct the plot result to.
#' Supported mechanisms: \code{"x11"} (default), \code{"png"}, \code{"jpg"},
#' \code{"bmp"}, \code{"pdf"} or \code{"ps"}.
#' @param path the path to create output files.
#' @param draw boolean to determine whether to plot the curves or just return the
#' calculated values (in cases where the user wants the output for later averaging
#' for example). Defaults to \code{TRUE} (make plots).
#' @param ... further arguments to be passed to plot devices, such as parameter
#' from \code{\link{par}}.
#' @return A named list with two members: the first member (\code{avg.ftdr})
#' contains a list with the means and the standard deviations of the averaged
#' \code{ftdr.obj} and are used to create the plot. The second member (\code{path})
#' contains the path to the created figure graphic.
#' @export
#' @author Panagiotis Moulos
#' @examples
#' \dontrun{
#' p11 <- 0.001*matrix(runif(300),100,3)
#' p12 <- matrix(runif(300),100,3)
#' p21 <- 0.001*matrix(runif(300),100,3)
#' p22 <- matrix(runif(300),100,3)
#' p31 <- 0.001*matrix(runif(300),100,3)
#' p32 <- matrix(runif(300),100,3)
#' p1 <- rbind(p11,p21)
#' p2 <- rbind(p12,p22)
#' p3 <- rbind(p31,p32)
#' rownames(p1) <- rownames(p2) <- rownames(p3) <-
#' paste("gene",1:200,sep="_")
#' colnames(p1) <- colnames(p2) <- colnames(p3) <-
#' paste("method",1:3,sep="_")
#' truth <- c(rep(1,40),rep(-1,40),rep(0,20),
#' rep(1,10),rep(2,10),rep(0,80))
#' names(truth) <- rownames(p1)
#' ftd.obj.1 <- diagplot.ftd(truth,p1,N=100,draw=FALSE)
#' ftd.obj.2 <- diagplot.ftd(truth,p2,N=100,draw=FALSE)
#' ftd.obj.3 <- diagplot.ftd(truth,p3,N=100,draw=FALSE)
#' ftd.obj <- list(ftd.obj.1,ftd.obj.2,ftd.obj.3)
#' avg.ftd.obj <- diagplot.avg.ftd(ftd.obj)
#'}
diagplot.avg.ftd <- function(ftdr.obj,output="x11",path=NULL,draw=TRUE,...) {
y.name <- list(
tpc="Number of True Positives",
fpc="Number of False Positives",
tnc="Number of True Negatives",
fnc="Number of False Negatives"
)
stats <- names(ftdr.obj[[1]]$ftdr)
type <- ftdr.obj[[1]]$type
truth <- ftdr.obj[[1]]$truth
N <- ftdr.obj[[1]]$N
avg.ftdr.obj <- vector("list",length(stats))
names(avg.ftdr.obj) <- stats
colspace.universe <- c("red","blue","green","orange","darkgrey","green4",
"black","pink","brown","magenta","yellowgreen","pink4","seagreen4",
"darkcyan")
colspace <- colspace.universe[1:length(stats)]
names(colspace) <- stats
for (s in stats) {
disp("Retrieving ",s)
avg.ftdr.obj[[s]] <- do.call("cbind",lapply(ftdr.obj,
function(x) x$ftdr[[s]]))
}
disp("")
avg.ftdr.obj <- lapply(avg.ftdr.obj,function(x) {
mn <- apply(x,1,mean)
st <- apply(x,1,sd)
return(list(mean=mn,sd=st))
})
means <- do.call("cbind",lapply(avg.ftdr.obj,function(x) x$mean))
stds <- do.call("cbind",lapply(avg.ftdr.obj,function(x) x$sd))
if (draw) {
fil <- file.path(path,paste("AVG_FTDR_",type,".",output,sep=""))
if (output %in% c("pdf","ps","x11"))
graphics.open(output,fil,width=8,height=8)
else
graphics.open(output,fil,width=1024,height=1024,res=100)
xlim <- ylim <- c(1,N)
par(cex.axis=0.9,cex.main=1,cex.lab=0.9,font.lab=2,font.axis=2,pty="m",
lwd=1.5,lty=1)
plot.new()
switch(type,
fpc = {
plot.window(xlim,ylim,log="y")
axis(1,at=pretty(xlim,10))
axis(2)
for (n in colnames(means)) {
lines(means[,n],col=colspace[n],...)
}
grid()
title(main="Selected genes vs False Positives",
xlab="Number of selected genes",ylab=y.name[[type]])
graphics::legend(x="topleft",legend=colnames(means),
col=colspace,lty=1)
},
tpc = {
plot.window(xlim,ylim)
axis(1,at=pretty(xlim,10))
axis(2,at=pretty(ylim,10))
for (n in colnames(means)) {
lines(means[,n],col=colspace[n],...)
}
grid()
title(main="Selected genes vs True Positives",
xlab="Number of selected genes",ylab=y.name[[type]])
graphics::legend(x="bottomright",legend=colnames(means),
col=colspace,lty=1)
},
fnc = {
plot.window(xlim,ylim,log="y")
axis(1,at=pretty(xlim,10))
axis(2)
for (n in colnames(means)) {
lines(means[,n],col=colspace[n],...)
}
grid()
title(main="Selected genes vs False Negatives",
xlab="Number of selected genes",ylab=y.name[[type]])
graphics::legend(x="topleft",legend=colnames(means),
col=colspace,lty=1)
},
tnc = {
plot.window(xlim,ylim)
axis(1,at=pretty(xlim,10))
axis(2,at=pretty(ylim,10))
for (n in colnames(means)) {
lines(means[,n],col=colspace[n],...)
}
grid()
title(main="Selected genes vs True Negatives",
xlab="Number of selected genes",ylab=y.name[[type]])
graphics::legend(x="bottomright",legend=colnames(means),
col=colspace,lty=1)
}
)
graphics.close(output)
}
else
fil <- NULL
return(list(avg.ftdr=list(means=means,stds=stds),path=fil))
}
#' Open plotting device
#'
#' Wrapper function to open a plotting device. Internal use only.
#'
#' @param o the plotting device, see main metaseqr function
#' @param f a filename, if the plotting device requires it (e.g. \code{"pdf"})
#' @param ... further arguments to be passed to plot devices, such as parameter
#' from \code{\link{par}}.
#' @author Panagiotis Moulos
#' @examples
#' \dontrun{
#' graphics.open("pdf","test.pdf",width=12,height=12)
#'}
graphics.open <- function(o,f,...) {
if(!check.graphics.type(o))
stopwrap("Invalid graphics output type!")
if(check.graphics.file(o) && is.null(f))
stopwrap("Please specify an output file name for your plot")
switch(o,
x11 = { dev.new(...) },
pdf = { pdf(file=f,pointsize=10,...) },
ps = { postscript(file=f,pointsize=10,...) },
png = { png(filename=f,pointsize=12,...) },
jpg = { jpeg(filename=f,pointsize=12,quality=100,...) },
bmp = { bmp(filename=f,pointsize=12,...) },
tiff = { tiff(filename=f,pointsize=12,...) }
)
}
#' Close plotting device
#'
#' Wrapper function to close a plotting device. Internal use only.
#'
#' @param o the plotting device, see main metaseqr function
#' @author Panagiotis Moulos
#' @examples
#' \dontrun{
#' graphics.close("pdf")
#'}
graphics.close <- function(o) {
if (!is.element(o,c("x11","png","jpg","tiff","bmp","pdf","ps")))
return(FALSE)
if (o!="x11")
dev.off()
}
#' Check plotting device
#'
#' Plotting device checker. Internal use only.
#'
#' @param o the plotting device, see main metaseqr function
#' @author Panagiotis Moulos
check.graphics.type <- function(o) {
if (!is.element(o,c("x11","png","jpg","tiff","bmp","pdf","ps")))
return(FALSE)
else
return(TRUE)
}
#' Check graphics file
#'
#' Graphics file checker. Internal use only.
#'
#' @param o the plotting device, see main metaseqr function
#' @author Panagiotis Moulos
check.graphics.file <- function(o) {
if (is.element(o,c("png","jpg","tiff","bmp","pdf","ps")))
return(TRUE)
else
return(FALSE)
}
#' Display value transformation
#'
#' Logarithmic transformation for display purposes. Internal use only.
#'
#' @param mat input data matrix
#' @param base logarithmic base, 2 or 10
#' @author Panagiotis Moulos
log2disp <- function(mat,base=2) {
mat[mat==0] <- 1
if (base==10)
return(log10(mat))
else
return(log2(mat))
}
#' General value transformation
#'
#' Logarithmic transformation. Internal use only.
#'
#' @param x input data matrix
#' @param base logarithmic base, 2 or 10
#' @param off offset to avoid Infinity
#' @author Panagiotis Moulos
nat2log <- function(x,base=2,off=1) {
#x[x==0] <- off
x <- x + off
if (base==2)
return(log2(x))
else
return(log10(x))
}
#' Old functions from NOISeq
#'
#' Old functions from NOISeq to create the \code{"readnoise"} plots. Internal use
#' only.
#'
#' @param input input to cddat.
#' @return a list with data to plot.
#' @note Adopted from an older version of NOISeq package (author: Sonia Tarazona).
#' @author Panagiotis Moulos
cddat <- function (input) {
if (inherits(input,"eSet") == FALSE)
stopwrap("The input data must be an eSet object.\n")
if (!is.null(assayData(input)$exprs)) {
if (ncol(assayData(input)$exprs) < 2)
stopwrap("The input object should have at least two samples.\n")
datos <- assayData(input)$exprs
}
else {
if (ncol(assayData(input)$counts) < 2)
stopwrap("The input object should have at least two samples.\n")
datos <- assayData(input)$counts
}
datos <- datos[which(rowSums(datos) > 0),]
nu <- nrow(datos) # number of detected features
qq <- 1:nu
data2plot = data.frame("%features" = 100*qq/nu)
for (i in 1:ncol(datos)) {
acumu <- 100*cumsum(sort(datos[,i],decreasing=TRUE))/sum(datos[,i])
data2plot = cbind(data2plot, acumu)
}
colnames(data2plot)[-1] = colnames(datos)
# Diagnostic test
KSpval = mostres = NULL
for (i in 1:(ncol(datos)-1)) {
for (j in (i+1):ncol(datos)) {
mostres = c(mostres, paste(colnames(datos)[c(i,j)], collapse="_"))
KSpval = c(KSpval, suppressWarnings(ks.test(datos[,i], datos[,j],
alternative = "two.sided"))$"p.value")
}
}
KSpval = p.adjust(KSpval, method = "fdr")
return(list(
"data2plot"=data2plot,
"DiagnosticTest"=data.frame("ComparedSamples"=mostres,"KSpvalue"=KSpval)
))
}
#' Old functions from NOISeq
#'
#' Old functions from NOISeq to create the \code{"readnoise"} plots. Internal use
#' only.
#' @param dat the returned list from \code{\link{cddat}}.
#' @param samples the samples to plot.
#' @param ... further arguments passed to e.g. \code{\link{par}}.
#' @return Nothing, it created the old RNA composition plot.
#' @note Adopted from an older version of NOISeq package (author: Sonia Tarazona)
#' @author Panagiotis Moulos
cdplot <- function (dat,samples=NULL,...) {
dat = dat$data2plot
if (is.null(samples)) samples <- 1:(ncol(dat)-1)
if (is.numeric(samples)) samples = colnames(dat)[samples+1]
colspace <- c("red","blue","yellowgreen","orange","aquamarine2","pink2",
"seagreen4","brown","purple","chocolate","gray10","gray30","darkblue",
"darkgreen","firebrick2","darkorange4","darkorchid","darkcyan","gold4",
"deeppink3")
if (length(samples)>length(colspace))
miscolores <- sample(colspace,length(samples),replace=TRUE)
else
miscolores <- sample(colspace,length(samples))
plot(dat[,1],dat[,samples[1]],xlab="% features",ylab="% reads",type="l",
col=miscolores[1],...)
for (i in 2:length(samples))
lines(dat[,1],dat[,samples[i]],col=miscolores[i])
graphics::legend("bottom",legend=samples,
text.col=miscolores[1:length(samples)],bty="n",lty=1,lwd=2,
col=miscolores[1:length(samples)])
}
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
Embedding an R snippet on your website
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.