Mercurial > repos > iuc > raceid_inspectclusters
view scripts/cluster.R @ 0:9fec5dd8fbb9 draft
planemo upload for repository https://github.com/galaxyproject/tools-iuc/tree/master/tools/raceid3 commit f880060c478d42202df5b78a81329f8af56b1138
author | iuc |
---|---|
date | Thu, 22 Nov 2018 04:45:41 -0500 |
parents | |
children | 64c5c1bbbdbe |
line wrap: on
line source
#!/usr/bin/env R VERSION = "0.2" args = commandArgs(trailingOnly = T) if (length(args) != 1){ message(paste("VERSION:", VERSION)) stop("Please provide the config file") } suppressWarnings(suppressPackageStartupMessages(require(RaceID))) suppressWarnings(suppressPackageStartupMessages(require(scran))) source(args[1]) do.filter <- function(sc){ if (!is.null(filt.lbatch.regexes)){ lar <- filt.lbatch.regexes nn <- colnames(sc@expdata) filt$LBatch <- lapply(1:length(lar), function(m){ return( nn[grep(lar[[m]], nn)] ) }) } sc <- do.call(filterdata, c(sc, filt)) ## Get histogram metrics for library size and number of features raw.lib <- log(colSums(as.matrix(sc@expdata))) raw.feat <- log(rowSums(as.matrix(sc@expdata))) filt.lib <- log(colSums(getfdata(sc))) filt.feat <- log(rowSums(getfdata(sc))) br <- 50 ## Determine limits on plots based on the unfiltered data ## (doesn't work, R rejects limits and norm data is too different to compare to exp data ## so let them keep their own ranges) ## betterrange <- function(floatval){ ## return(10 * (floor(floatval / 10) + 1)) ## } ## tmp.lib <- hist(raw.lib, breaks=br, plot=F) ## tmp.feat <- hist(raw.feat, breaks=br, plot=F) ## lib.y_lim <- c(0,betterrange(max(tmp.lib$counts))) ## lib.x_lim <- c(0,betterrange(max(tmp.lib$breaks))) ## feat.y_lim <- c(0,betterrange(max(tmp.feat$counts))) ## feat.x_lim <- c(0,betterrange(max(tmp.feat$breaks))) par(mfrow=c(2,2)) print(hist(raw.lib, breaks=br, main="ExpData Log(LibSize)")) # , xlim=lib.x_lim, ylim=lib.y_lim) print(hist(raw.feat, breaks=br, main="ExpData Log(NumFeat)")) #, xlim=feat.x_lim, ylim=feat.y_lim) print(hist(filt.lib, breaks=br, main="FiltData Log(LibSize)")) # , xlim=lib.x_lim, ylim=lib.y_lim) print(hist(filt.feat, breaks=br, main="FiltData Log(NumFeat)")) # , xlim=feat.x_lim, ylim=feat.y_lim) if (filt.use.ccorrect){ par(mfrow=c(2,2)) sc <- do.call(CCcorrect, c(sc, filt.ccc)) print(plotdimsat(sc, change=T)) print(plotdimsat(sc, change=F)) } return(sc) } do.cluster <- function(sc){ sc <- do.call(compdist, c(sc, clust.compdist)) sc <- do.call(clustexp, c(sc, clust.clustexp)) if (clust.clustexp$sat){ print(plotsaturation(sc, disp=F)) print(plotsaturation(sc, disp=T)) } print(plotjaccard(sc)) return(sc) } do.outlier <- function(sc){ sc <- do.call(findoutliers, c(sc, outlier.findoutliers)) if (outlier.use.randomforest){ sc <- do.call(rfcorrect, c(sc, outlier.rfcorrect)) } print(plotbackground(sc)) print(plotsensitivity(sc)) print(plotoutlierprobs(sc)) ## Heatmaps test1 <- list() test1$side = 3 test1$line = 0 #1 #3 x <- clustheatmap(sc, final=FALSE) print(do.call(mtext, c(paste("(Initial)"), test1))) ## spacing is a hack x <- clustheatmap(sc, final=TRUE) print(do.call(mtext, c(paste("(Final)"), test1))) ## spacing is a hack return(sc) } do.clustmap <- function(sc){ sc <- do.call(comptsne, c(sc, cluster.comptsne)) sc <- do.call(compfr, c(sc, cluster.compfr)) return(sc) } mkgenelist <- function(sc){ ## Layout test <- list() test$side = 3 test$line = 0 #1 #3 test$cex = 0.8 df <- c() options(cex = 1) lapply(unique(sc@cpart), function(n){ dg <- clustdiffgenes(sc, cl=n, pvalue=genelist.pvalue) dg.goi <- dg[dg$fc > genelist.foldchange,] dg.goi.table <- head(dg.goi, genelist.tablelim) df <<- rbind(df, cbind(n, dg.goi.table)) goi <- head(rownames(dg.goi.table), genelist.plotlim) print(plotmarkergenes(sc, goi)) print(do.call(mtext, c(paste(" Cluster ",n), test))) ## spacing is a hack test$line=-1 print(do.call(mtext, c(paste(" Sig. Genes"), test))) ## spacing is a hack test$line=-2 print(do.call(mtext, c(paste(" (fc > ", genelist.foldchange,")"), test))) ## spacing is a hack }) write.table(df, file=out.genelist, sep="\t", quote=F) } pdf(out.pdf) if (use.filtnormconf){ sc <- do.filter(sc) message(paste(" - Source:: genes:",nrow(sc@expdata),", cells:",ncol(sc@expdata))) message(paste(" - Filter:: genes:",nrow(sc@ndata),", cells:",ncol(sc@ndata))) message(paste(" :: ", sprintf("%.1f", 100 * nrow(sc@ndata)/nrow(sc@expdata)), "% of genes remain,", sprintf("%.1f", 100 * ncol(sc@ndata)/ncol(sc@expdata)), "% of cells remain")) } if (use.cluster){ par(mfrow=c(2,2)) sc <- do.cluster(sc) par(mfrow=c(2,2)) sc <- do.outlier(sc) par(mfrow=c(2,2), mar=c(1,1,6,1)) sc <- do.clustmap(sc) mkgenelist(sc) } dev.off() saveRDS(sc, out.rdat)