Nothing
#' Get overlap information for pairs of gene knock-outs from LINCS data
#' @export
#' @importFrom utils combn data
#' @import hgu133a.db
#' @import KOdata
#' @param KGML_file An object of formal class KEGGPathway
#' @param KEGG_mappings The data.frame object generated by the function
#' expand_KEGG_mappings
#' @param cell_type Choose from the set of cell lines:
#' (A375,A549,ASC,HA1E,HCC515,HEK293T,HEKTE,HEPG2,HT29,MCF7,NCIH716,NPC,PC3,
#' SHSY5Y,SKL,SW480,VCAP)
#' @param data_type Choose from data types: (100_full, 100_bing, 50_lm)
#' @param pert_time Choose from (6,24,48,96,120,144,168)
#' @param only_mapped A logical indicator; if set to FALSE will return 'de-novo'
#' edges that 'exist' in data but are not documented in KEGG
#' @param affy_based A logical indicator; if set to TRUE will return
#' lists/counts based on probeID instead of gene symbol.
#' @param keep_counts_only A logical indicator; if set to FALSE will return
#' data frame with lists [of gene symbols or probe ids] as well as counts
#' @param add_fisher_information A logical indicator; by default the
#' relationships are analyzed for strength of correlation via
#' Fisher's Exact Test
#' @param p.adjust.method For available methods, type 'p.adjust.methods' into
#' command promt and press enter.
#' @return A data frame where each row corresponds to information for pairs of
#' experimental gene knock-outs from LINCS data (found in selected pathway).
#' @examples
#' p53_KGML <- get_KGML("hsa04115")
#' p53_KEGG_mappings <- expand_KEGG_mappings(p53_KGML)
#' p53_edges <- expand_KEGG_edges(p53_KGML, p53_KEGG_mappings)
#'
#' summary <- path_genes_by_cell_type(p53_KEGG_mappings)
#' p53_HA1E_data <- overlap_info(p53_KGML, p53_KEGG_mappings,
#' "HA1E", data_type = "100_bing",
#' only_mapped = FALSE)
overlap_info <-
function(KGML_file, KEGG_mappings, cell_type,
data_type = "100_full",pert_time = 96,
only_mapped = TRUE, affy_based = FALSE,
keep_counts_only = TRUE,
add_fisher_information = TRUE,
p.adjust.method = "BH"){
data("KO_data", envir = environment())
KO_data <- get("KO_data")
if(!cell_type %in% unique(KO_data$cell_id)){
warning("Knock-out data is not available for selected cell-line \n")
overlaps <- data.frame("null" = NA)
return(overlaps)
}
KO_by_CT <- KO_data[KO_data$pert_time == pert_time &
KO_data$cell_id == cell_type,]
expanded_edges <- expand_KEGG_edges(KGML_file, KEGG_mappings)
keeps <- c("entry1symbol", "entry2symbol")
expanded_edges <- expanded_edges[expanded_edges$type != "maplink",keeps]
if(nrow(expanded_edges) == 0 & only_mapped){
warning("No documented edges are found in data;
only data for de-novo edges can be generated \n")
overlaps <- data.frame("null" = NA)
return(overlaps)
}
keeps <- c("entryACCESSION", "entrySYMBOL")
path_genes <- KEGG_mappings[KEGG_mappings$entryTYPE == "gene", keeps]
names(path_genes)[2] <- "SYMBOL"
path_genes <- data.frame("ENTREZID" = unlist(path_genes$entryACCESSION),
"SYMBOL" = unlist(path_genes$SYMBOL),
stringsAsFactors = FALSE)
path_genes <- path_genes[!duplicated(path_genes$ENTREZID),]
data <- subset(path_genes, path_genes$SYMBOL %in% KO_by_CT$pert_desc)
cat(paste0("Number of genes documented in selected pathway = ",
nrow(path_genes)), "\n")
cat(paste0("Number of pathway genes in dataset = ", nrow(data),"\n"))
cat(paste0("Coverage = ", round(nrow(data)/nrow(path_genes),4)*100,
"%","\n"))
if(nrow(data) == 0 | nrow(data) == 1){
warning("Overlap data for selected pathway cannot be generated for
selected cell line")
overlaps <- data.frame("null" = NA)
return(overlaps)
}
conversion_key <- suppressWarnings(suppressMessages(
AnnotationDbi::select(hgu133a.db::hgu133a.db,
AnnotationDbi::keys(hgu133a.db), c("SYMBOL","ENTREZID"), "PROBEID")))
for (i in 1:nrow(data)){
if (data_type == "100_full"){
data$up[i] <-
strsplit(KO_by_CT$up100_full[KO_by_CT$pert_desc ==
data$SYMBOL[i]], ";")
data$down[i] <-
strsplit(KO_by_CT$dn100_full[KO_by_CT$pert_desc ==
data$SYMBOL[i]], ";")
}
else if (data_type == "100_bing"){
data$up[i] <-
strsplit(KO_by_CT$up100_bing[KO_by_CT$pert_desc ==
data$SYMBOL[i]], ";")
data$down[i] <-
strsplit(KO_by_CT$dn100_bing[KO_by_CT$pert_desc ==
data$SYMBOL[i]], ";")
}
else if (data_type == "50_lm"){
data$up[i] <-
strsplit(KO_by_CT$up50_lm[KO_by_CT$pert_desc ==
data$SYMBOL[i]], ";")
data$down[i] <-
strsplit(KO_by_CT$dn50_lm[KO_by_CT$pert_desc ==
data$SYMBOL[i]], ";")
}
else {
warning("valid data_type options: 100_full, 100_bing, 50_lm")
return(NA)
}
data$up_symbol[i] <-
list(conversion_key$SYMBOL[which(conversion_key$PROBEID %in%
unlist(data$up[i]))])
data$down_symbol[i] <-
list(conversion_key$SYMBOL[which(conversion_key$PROBEID %in%
unlist(data$down[i]))])
}
overlaps<- data.frame(t(combn(data$SYMBOL,2)), stringsAsFactors = FALSE)
names(overlaps) <- c("knockout1", "knockout2")
overlaps$unique_ID <- paste0(overlaps$knockout1, ",", overlaps$knockout2)
expanded_edges$unique_ID <- paste0(expanded_edges$entry1symbol, ",",
expanded_edges$entry2symbol)
expanded_edges$unique_IDR <- paste0(expanded_edges$entry2symbol, ",",
expanded_edges$entry1symbol)
pre_mapped1 <- subset(overlaps, overlaps$unique_ID %in%
expanded_edges$unique_ID) ## Direction is correct
pre_mapped2 <- subset(overlaps, overlaps$unique_ID %in%
expanded_edges$unique_IDR) ## Direction is reversed
pre_mapped2 <- pre_mapped2[,c(2,1)]
names(pre_mapped2)[c(1,2)] = names(pre_mapped1)[c(1,2)]
if(nrow(pre_mapped2) >= 1 & nrow(pre_mapped1) >= 1){
pre_mapped2$unique_ID <- paste0(pre_mapped2$knockout1, ",",
pre_mapped2$knockout2)
pre_mapped <- rbind(pre_mapped1, pre_mapped2)
}
if (nrow(pre_mapped2) == 0 & (nrow(pre_mapped1) == 0)){
pre_mapped <- data.frame("unique_ID" = NA)
}
if (nrow(pre_mapped1) == 0 & nrow(pre_mapped2) >= 1){
pre_mapped2$unique_ID <- paste0(pre_mapped2$knockout1, ",",
pre_mapped2$knockout2)
pre_mapped <- pre_mapped2
}
if (nrow(pre_mapped2) == 0 & nrow(pre_mapped1) >= 1){
pre_mapped <- pre_mapped1
}
if(is.na(pre_mapped[1,1]) & only_mapped){
warning("No documented edges are found in data; only data for
de-novo edges can be generated \n")
overlaps <- data.frame("null" = NA)
return(overlaps)
}
pre_mapped$pre_mapped <- 1
keeps <-c("knockout1", "knockout2", "pre_mapped")
pre_mapped <- pre_mapped[,keeps]
if (!only_mapped){
un_mapped <- subset(overlaps, !overlaps$unique_ID %in%
expanded_edges$unique_ID & !overlaps$unique_ID %in%
expanded_edges$unique_IDR)
un_mapped$pre_mapped <- 0
un_mapped <- un_mapped[,keeps]
overlaps <- rbind(pre_mapped, un_mapped)
}
else {
overlaps <- pre_mapped
}
for (i in 1:nrow(overlaps)){
overlaps$affy.genesUP[i] <-
list(intersect(unlist(data$up[which(data$SYMBOL == overlaps[i,1])]),
unlist(data$up[which(data$SYMBOL == overlaps[i,2])])))
overlaps$affy.genesDOWN[i] <-
list(intersect(unlist(data$down[which(data$SYMBOL == overlaps[i,1])]),
unlist(data$down[which(data$SYMBOL == overlaps[i,2])])))
overlaps$affy.genesUPK1_DOWNK2[i] <-
list(intersect(unlist(data$up[which(data$SYMBOL == overlaps[i,1])]),
unlist(data$down[which(data$SYMBOL == overlaps[i,2])])))
overlaps$affy.genesDOWNK1_UPK2[i] <-
list(intersect(unlist(data$down[which(data$SYMBOL == overlaps[i,1])]),
unlist(data$up[which(data$SYMBOL == overlaps[i,2])])))
overlaps$num.affy.genesUP[i] <- length(unlist(overlaps$affy.genesUP[i]))
overlaps$num.affy.genesDOWN[i] <- length(unlist(overlaps$affy.genesDOWN[i]))
overlaps$num.affy.genesUPK1_DOWNK2[i] <-
length(unlist(overlaps$affy.genesUPK1_DOWNK2[i]))
overlaps$num.affy.genesDOWNK1_UPK2[i] <-
length(unlist(overlaps$affy.genesDOWNK1_UPK2[i]))
}
if (!affy_based){
for(i in 1:nrow(overlaps)){
overlaps$genes.symbolsUP[i]<-
list(conversion_key$SYMBOL[which(conversion_key$PROBEID %in%
unlist(overlaps$affy.genesUP[i]))])
overlaps$genes.symbolsDOWN[i]<-
list(conversion_key$SYMBOL[which(conversion_key$PROBEID %in%
unlist(overlaps$affy.genesDOWN[i]))])
overlaps$genes.symbolsUPK1_DOWNK2[i]<-
list(conversion_key$SYMBOL[which(conversion_key$PROBEID %in%
unlist(overlaps$affy.genesUPK1_DOWNK2[i]))])
overlaps$genes.symbolsDOWNK1_UPK2[i]<-
list(conversion_key$SYMBOL[which(conversion_key$PROBEID %in%
unlist(overlaps$affy.genesDOWNK1_UPK2[i]))])
overlaps$num.genes.symbolsUP[i] <-
length(unlist(overlaps$genes.symbolsUP[i]))
overlaps$num.genes.symbolsDOWN[i] <-
length(unlist(overlaps$genes.symbolsDOWN[i]))
overlaps$num.genes.symbolsUPK1_DOWNK2[i] <-
length(unlist(overlaps$genes.symbolsUPK1_DOWNK2[i]))
overlaps$num.genes.symbolsDOWNK1_UPK2[i] <-
length(unlist(overlaps$genes.symbolsDOWNK1_UPK2[i]))
}
keeps = c("knockout1","knockout2","num.genes.symbolsUP",
"num.genes.symbolsDOWN", "num.genes.symbolsUPK1_DOWNK2",
"num.genes.symbolsDOWNK1_UPK2", "genes.symbolsUP",
"genes.symbolsDOWN", "genes.symbolsUPK1_DOWNK2",
"genes.symbolsDOWNK1_UPK2", "pre_mapped")
overlaps <- overlaps[,keeps]
}
else {
keeps = c("knockout1","knockout2","num.affy.genesUP",
"num.affy.genesDOWN", "num.affy.genesUPK1_DOWNK2",
"num.affy.genesDOWNK1_UPK2", "affy.genesUP", "affy.genesDOWN",
"affy.genesUPK1_DOWNK2", "affy.genesDOWNK1_UPK2", "pre_mapped")
overlaps <- overlaps[, keeps]
}
names(overlaps)[c(3:6)] <- c("UP", "DOWN", "UK1_DK2", "DK1_UK2")
method = p.adjust.method
if (keep_counts_only){
keeps = c("knockout1","knockout2", "UP", "DOWN", "UK1_DK2", "DK1_UK2",
"pre_mapped")
overlaps <- overlaps[,keeps]
}
if(add_fisher_information){
overlaps <- get_fisher_info(overlaps, method)
}
return(overlaps)
}
Any scripts or data that you put into this service are public.
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.