R/dynRangePlot.R

In erccdashboard: Assess Differential Gene Expression Experiments with ERCC Controls

#' Produce signal-abundance plot to evaluate dynamic range
#'
#' @param exDat      list, contains input data and stores analysis results
#' @param allPoints  boolean, default is false, means of replicates will be 
#'                      plotted. If true then all replicates will be plotted as
#'                      individual points.
#' @param labelReps  boolean, default is false. If true then replicates will be
#'                      labeled.
#'                      
#' @examples
#' data(SEQC.Example)
#' 
#' exDat <- initDat(datType="count", isNorm=FALSE, exTable=MET.CTL.countDat, 
#'                  filenameRoot="testRun", sample1Name="MET", 
#'                  sample2Name="CTL", erccmix="RatioPair", 
#'                  erccdilution=1/100, spikeVol=1, totalRNAmass=0.500,
#'                  choseFDR=0.1)
#'                  
#' exDat <- est_r_m(exDat)
#'                   
#' exDat <- dynRangePlot(exDat, allPoints="FALSE", labelReps ="FALSE")
#' 
#' exDat$Figures$dynRangePlot
#' 
#' @export
dynRangePlot <- function(exDat, allPoints, labelReps){
    
    ## Assign local variables
    sampleInfo <- exDat$sampleInfo
    plotInfo <- exDat$plotInfo
    erccInfo <- exDat$erccInfo
    
    ## check for allPoints argument
    if(missing(allPoints)){
        allPoints <- FALSE
    }
    ## check for labelReps arguement
    if(missing(labelReps)){
        labelReps <- FALSE
    }
    
    if(is.null(exDat$idColsAdj)){
        idCols <- exDat$idCols
        cat("\nmRNA fraction estimate is assumed to be equal to 1")
    }else{
        idCols <- exDat$idColsAdj    
    }
    
    sampleNames <- exDat$sampleNames
    indivxlabel <- plotInfo$ERCCxlabelIndiv
    avexlabel<- plotInfo$ERCCxlabelAve
    
    expressDat <- exDat$normERCCDat
    
    totCol <- ncol(expressDat[-c(1)])
    if(odd(totCol)) stop("Uneven number of replicates for the two sample types")
    
    # Get number of replicates
    reps <- (ncol(expressDat[-c(1)]))/2

    # Default is to plot errorbars
    errorBars <- TRUE
    # If there are not at least 3 reps, don't plot errorbars
    
    if (reps < 3){
        errorBars <- FALSE
        cat(paste0("\nLess than 3 replicates per sample, no sd values ",
                   "or error bars will be available.\n"))
    }
    if (allPoints == TRUE){
        errorBars <- FALSE
        cat(paste0("\nallPoints == TRUE, every replicate point will be plotted.\n"))
        if(labelReps == TRUE){
            cat(paste0("\nlabelReps == TRUE, points will be labeled.\n"))
        }
    }
    
    designMat <- exDat$designMat
    FCcode <- erccInfo$FCcode 
    
    myXLim <- plotInfo$myXLim 
    myYLim <- plotInfo$myYLim
    xlimEffects <- plotInfo$xlimEffects
    filenameRoot <- sampleInfo$filenameRoot
    
    colScale <- plotInfo$colScale
    fillScale <- plotInfo$fillScale
    
    # Create dynamic range data frame for ERCC Mix 1 samples
    dynRangeDatMix1 <- merge(idCols[c(1, 4, 5)], expressDat)
    names(dynRangeDatMix1)[3] <- "Conc"
    cat(paste("\nNumber of ERCCs in Mix 1 dyn range: ",
              dim(dynRangeDatMix1)[1], "\n"))
    
    # Melt the data to create long data frame for ggplot
    dynRangeDatMix1_l <- reshape2::melt(dynRangeDatMix1,
                             id.vars=c("Feature", "Ratio", "Conc") ) 
    
    # Create data.frame of the experimental design factors
    designVar <- reshape2::colsplit(dynRangeDatMix1_l$variable, pattern="_",
                         names=names(designMat)[-1])
    designVar <- as.data.frame(lapply(designVar, as.factor))  
    
    # Bind Mix 1 data and experimental design factors
    dynRangeDatMix1_l <- cbind(dynRangeDatMix1_l, designVar)
    
    #Create dynamic range data frame for ERCC Mix 2 samples
    dynRangeDatMix2 <- merge(idCols[c(1, 4, 6)], expressDat)
    names(dynRangeDatMix2)[3] <- "Conc"
    cat(paste("\nNumber of ERCCs in Mix 2 dyn range: ",
              dim(dynRangeDatMix2)[1], "\n"))
    
    # Melt the data to create long data frame for ggplot
    dynRangeDatMix2_l <- reshape2::melt(dynRangeDatMix2,
                              id.vars=c("Feature", "Ratio", "Conc") )
    
    # Create data.frame of the experimental design factors
    designVar <- reshape2::colsplit(dynRangeDatMix2_l$variable, pattern="_", 
                          names=names(designMat)[-1])
    designVar <- as.data.frame(lapply(designVar, as.factor))
    
    # Bind Mix 2 data and experimental design factors
    dynRangeDatMix2_l <- cbind(dynRangeDatMix2_l, designVar)
    
    # Sort the data
    dynRangeDatMix1_l <- dynRangeDatMix1_l[do.call(order, 
                                                  dynRangeDatMix1_l[c(3)]),]  
    dynRangeDatMix2_l <- dynRangeDatMix2_l[do.call(order, 
                                                  dynRangeDatMix2_l[c(3)]),]
    
    # Merge the dynamic range data frames
    dynRangeDatMix1_l$MixConc <- "Mix1"
    dynRangeDatMix2_l$MixConc <- "Mix2"
    dynRangeDatMix1 <- subset(dynRangeDatMix1_l, (Sample == sampleNames[1]))
    dynRangeDatMix2 <- subset(dynRangeDatMix2_l, (Sample == sampleNames[2]))
    
    dynRangeDat_l <- rbind(dynRangeDatMix1, dynRangeDatMix2)
    dynRangeDat_l$Feature <- as.factor(as.character(dynRangeDat_l$Feature))
    
    #browser()
    
    #####
    # Signal Abundance of Plot Sample A and B
    #####
    
    AandB <- subset(dynRangeDat_l,
                    ((Sample == sampleNames[1])|(Sample == sampleNames[2])))
    Adat <- subset(AandB, Sample == sampleNames[1])
    Bdat <- subset(AandB, Sample == sampleNames[2])
    
    # Log the data and then filter out NA and Inf Mn and Sd values
    Adat$value <- log2(Adat$value)
    Adat$Conc <- log2(Adat$Conc)
    
    AdatAve <- data.frame(tapply(Adat$value, Adat[,"Feature"], mean))
    AdatSD <- as.vector(tapply(Adat$value, Adat[,"Feature"], sd))
    AdatAveSD <- cbind(row.names(AdatAve), Adat$Ratio[match(row.names(AdatAve),
                                                          table=Adat$Feature)],
                      Adat$Sample[match(row.names(AdatAve), 
                                        table=Adat$Feature)],
                      Adat$Conc[match(row.names(AdatAve), table=Adat$Feature)],
                      AdatAve[c(1)])
    names(AdatAveSD)[1:5] <- c("Feature", "Ratio", "Sample", "Conc",
                               "value.Ave")
    
    AdatAveSD$value.SD <- AdatSD
    
    Bdat$value <- log2(Bdat$value)
    Bdat$Conc <- log2(Bdat$Conc)
    
    BdatAve <- data.frame(tapply(Bdat$value, Bdat[,"Feature"], mean))
    BdatSD <- as.vector(tapply(Bdat$value, Bdat[,"Feature"], sd))
    
    BdatAveSD <- cbind(row.names(BdatAve), Bdat$Ratio[match(row.names(BdatAve), 
                                                          table=Bdat$Feature)], 
                      Bdat$Sample[match(row.names(BdatAve), 
                                        table=Bdat$Feature)],
                      Bdat$Conc[match(row.names(BdatAve), table=Bdat$Feature)], 
                      BdatAve[c(1)])
    names(BdatAveSD)[1:5] <- c("Feature", "Ratio", "Sample", "Conc", 
                               "value.Ave")
    
    BdatAveSD$value.SD <- BdatSD
    
    AandBAveSD <- rbind(AdatAveSD, BdatAveSD)
    cutERCCs <- unique(AandBAveSD$Feature[which(is.na(AandBAveSD$value.SD))])
    
    #print(which(AandBAveSD$Feature %in% cutERCCs))
    if(length(cutERCCs) != 0){
        cat(paste("These ERCCs were not included in the signal-abundance plot,", 
                  "because not enough non-zero replicate measurements of these",
                  " controls were obtained for both samples:\n", "", sep='\n'))
        cutERCCs <- as.character(cutERCCs)
        for (j in seq(from=1, to=length(cutERCCs), by=5)){
            k <- j+4
            if (k > length(cutERCCs)) k <- length(cutERCCs)
            cat(cutERCCs[j:k])
            cat("\n")
        } 
        AandBAveSD <- AandBAveSD[-(which(AandBAveSD$Feature %in% cutERCCs)),]  
    } 
    
    AandBAveSD$Feature <- as.factor(as.character(AandBAveSD$Feature))
    
    AandBAveSD$value.Ave <- as.vector(AandBAveSD$value.Ave)
    
    #set limits and axis labels
    if(sampleInfo$datType == "count"){
        yinfo <- ylab("Log2 Normalized ERCC Counts")
    }
    if(sampleInfo$datType == "array"){
        yinfo <- ylab("Log2 Normalized Fluorescent Intensity")
    }
    
    xlabel <- xlab(indivxlabel)
    
    conc <- NULL
    idColsOrig <- exDat$idCols
    for (i in grep("Conc", names(idColsOrig))){
        conc <- c(conc, idColsOrig[,i])  
    }
    xmin <- min(log2(conc)) - 1
    xmax <- max(log2(conc)) + 1
    myXLim <- c(xmin, xmax)
    
    if(is.null(myYLim)){
        ymin <- min(log2(AandB$value)) - 1
        ymax <- max(log2(AandB$value)) +1
        myYLim <- c(ymin, ymax)
    }
    if((sampleInfo$datType == "array")&
           (myYLim[1]+1 < min(log2(AandB$value)))){
        ymin <- min(log2(AandB$value)) - 1
        ymax <- max(log2(AandB$value)) +1
        myYLim <- c(ymin, ymax)
    }
    myYLim <- myYLim
    myXLim <- myXLim
    exDat$plotInfo$myXLim <- myXLim
    exDat$plotInfo$myYLim <- myYLim
    
    plotLim <- coord_cartesian(xlim=myXLim, ylim=myYLim)
    
    ### Create a linear model for the data
    dataFit <- AandBAveSD
    #write.csv(dataFit, paste(filenameRoot, "dataFit.csv"))
    # Create smoothed variance weights based on spline
    
    data.lo <- loess(value.SD~Conc, data=dataFit)
    (data.lo)
    loessWtEst <- ggplot(dataFit, aes(x=Conc, y=value.SD)) + 
        geom_point() + 
        geom_smooth(method="loess")
    
    
    data.lo.pred <- predict(data.lo, dataFit$Conc) 
    
    wtSet <- 1/((data.lo.pred)^2)
    
    dataFit$wtSet <- wtSet
    
    # For case of reps < 3 set sd to 0, so no errorBars are plotted
    if(errorBars == FALSE) AandBAveSD$value.SD <- 0
    
    # Appease R CMD Check
    Sample <- Conc <- value.SD <- value.Ave <- Ratio <- ERCC.effect <- NULL
    Feature <- NULL
    
    if (allPoints == FALSE){
        # plot mean and sd error bars for dynamic range plot
        dynRange <- ggplot() + 
            geom_pointrange(data=subset(AandBAveSD, Sample == sampleNames[1]),
                            aes(x=Conc, y=value.Ave, ymax=value.Ave + value.SD, 
                                ymin=value.Ave - value.SD, colour=Ratio, 
                                shape=Sample), alpha=0.6, size=1.25) + 
            geom_pointrange(data=subset(AandBAveSD, Sample == sampleNames[2]),
                            aes(x=Conc, y=value.Ave, ymax=value.Ave + value.SD,
                                ymin=value.Ave - value.SD, colour=Ratio,
                                shape=Sample), alpha=0.6, size=1.25) + xlabel + 
            yinfo + plotLim + colScale + 
            theme_bw() +theme( legend.justification=c(0, 1), legend.position=c(0, 1))    
    }
    if (allPoints == TRUE){
        # include all data points on dynamic range plot
        if(labelReps == FALSE){
            dynRange <- ggplot(AandB) + 
                geom_point(aes(x = log2(Conc), y = log2(value), 
                               colour = Ratio, shape = Sample),
                           alpha = 0.6, size = 6) + 
                xlabel + 
                yinfo + plotLim + colScale + 
                theme_bw() +theme( legend.justification=c(0, 1), legend.position=c(0, 1))    
        }else{
            dynRange <- ggplot(AandB) + 
                geom_point(aes(x = log2(Conc), y = log2(value), 
                               colour = Ratio, shape = Sample),
                           alpha = 0.6, size = 6) + 
                geom_text(aes(x = log2(Conc), y = log2(value), label = Rep )) +
                xlabel + 
                yinfo + plotLim + colScale + 
                theme_bw() +theme( legend.justification=c(0, 1), legend.position=c(0, 1))
            
        }
            
    }
    
    
    ### Serial model approach
    
    ## fit1 is linear model of value.Ave and Conc, needed for the LODR to 
    # Concentration conversion
    # fit2 adds in ERCC to the fit1 model
    ## comparison of fit2 to fit1 should allow for t-test to identify outlier 
    # ERCC controls on the basis of a per ERCC slope and intercept comparison 
    # to the fit1 slope
    fit1.lm <- lm(data=dataFit, formula=value.Ave~Conc, weights=wtSet)
    fit2.lm <- lm(data=dataFit, formula=value.Ave~Conc+as.factor(Feature),
                 weights=wtSet)
    
    perERCC<-as.data.frame(coef(summary(fit2.lm)))
    #print("perERCC data frame")
    
    perERCC<- perERCC[-(2),]
    
    # compare ERCC specific intercepts to overall fit1 intercept
    perERCC[,1]<-(perERCC[1, 1]+c(0, perERCC[-1, 1]))-coef(fit1.lm)[1]
    
    Conc1 <- dataFit$Conc[which(dataFit$Sample == sampleNames[1])]
    Conc2 <- dataFit$Conc[which(dataFit$Sample == sampleNames[2])]
    AveConc <- (Conc1 + Conc2)/2
    
    ## standardized ERCC effect <- perERCC[,1]/perERCC[,2], consider it size of 
    ## ERCC effect in standard errors (unitless)
    effects <- data.frame(Feature=levels(dataFit$Feature), AveConc=AveConc,
                          ERCC.effect=perERCC[,1]/perERCC[,2])
    
    effects$Feature <- as.factor(effects$Feature)
    effects$Ratio <- dataFit$Ratio[1:nlevels(dataFit$Feature)]
    
    minLabel <- effects$ERCC.effect <= fivenum(effects$ERCC.effect)[2]- 
        1.5*IQR(effects$ERCC.effect)
    maxLabel <- effects$ERCC.effect >= fivenum(effects$ERCC.effect)[4]+ 
        1.5*IQR(effects$ERCC.effect)
    xlabel <- xlab(avexlabel)
    if(xlimEffects[1] > min(effects$ERCC.effect)){
        xlimEffects <- c((min(effects$ERCC.effect)),
                        (abs(min(effects$ERCC.effect)))) 
    }
    
    #if((any(minLabel))|(any(maxLabel))){
    if((any(minLabel))){
        effectsPlot <- ggplot(effects, aes(x=AveConc, y=ERCC.effect)) + 
            geom_point(aes(colour=Ratio), size=6, alpha=0.8) + xlabel +
            ylab(paste("Per ERCC Differences from Linear Fit (Standardized",
                       "by s.e., unitless)")) +
            coord_cartesian(ylim=xlimEffects, xlim=myXLim) + colScale + 
            geom_text(data=subset(effects,
                                  (ERCC.effect <= fivenum(ERCC.effect)[2] - 
                                               (1.5)*IQR(ERCC.effect))),
                      aes(x=AveConc, y=ERCC.effect, label=gsub("ERCC-00","",
                                                                     Feature)),
                      colour="black", show.legend=FALSE, angle=45, hjust=-0.25, 
                      position=position_jitter(width=0.5)) + theme_bw() + 
            theme(legend.justification=c(0, 0), legend.position=c(0, 0))
    }else{
        effectsPlot <- ggplot(effects, aes(x=AveConc, y=ERCC.effect)) + 
            geom_point(aes(colour=Ratio), size=6, alpha=0.8) + xlabel + 
            ylab(paste("Per ERCC Differences from Linear Fit (Standardized",
                       "by s.e., unitless)")) +
            coord_cartesian(ylim= xlimEffects, xlim=myXLim) + 
            colScale + 
            theme_bw() + 
            theme(legend.justification=c(0, 0), legend.position=c(0, 0)) 
    }  
    
    # Get the residuals from fit1.lm
    
    fit.coeff <- unlist(fit1.lm$coefficients)
    #exDat$fit.coeff <- fit.coeff
    
    
    if(sampleInfo$datType == "array"){
        effectsPlot <- NULL
        cat(paste("\ndatType is array, no linear model fit",
                  "for ERCC specific effects\n"))
        cat("\nrangeResidPlot is empty")
    }
    cat("\n\nSaving dynRangePlot to exDat\n")
    # save dynamic range plot
    exDat$Figures$dynRangePlot <- dynRange
    # save ERCC specific effects plot
    exDat$Figures$rangeResidPlot <- effectsPlot
    
    
    exDat$Results$dynRangeDat <- AandBAveSD
    exDat$Results$rangeResidDat <- effects
    return(exDat)
 
}