A foreach loop throws an error in running parallel tasks

I am running differential expression analysis code using edgeR written by authors of RSEQREP pipeline(opensource) , however the DEG step keeps throwing the following error.

I have managed to pinpoint the error originating from the foreach parallel loop part , but I don't know how to go about it, please help. The code is as shown below. I am using R version 3.6.0 and due to dependency issues its the only one I can work with and ubuntu 18.04

source('init-analysis.r')

## set directories
in.dir.lcpm  = paste(res.dir,'lcpm',sep='/');
out.dir.glm	= paste(res.dir,'glm',sep='/');

# parallel implementation
registerDoParallel(cores=ncores)
splits = split(expand.grid(1:length(spcFlags),1:length(trtFlags),1:length(postb.times)),1:(length(spcFlags)*length(trtFlags)*length(postb.times)))
foreach (z=1:length(splits)) %dopar% {
	split = unlist(splits[[z]])
	spcFlags = spcFlags[split[1]]
	trtFlags = trtFlags[split[2]]
	postb.times = postb.times[split[3]]
	
	## loop through specimen types;
	for(s in 1:length(spcFlags)) {
		spcFlag = spcFlags[s];
		spcLabl = spcFlags[s];
		
		## load edgeR DGE object
		load(file=paste(out.dir.glm,'/',spcFlag,'_alltp_edge_r_dge.RData',sep=''));
		
		## loop through treatment groups
		for (v in 1:length(trtFlags)) {
			trtFlag = trtFlags[v];
		
			## loop through each post treatment time
			for(t in 1:length(postb.times)) {
				time = postb.times[t];
				dge.lab.time=dge.spc;
			
				## select subjects that have both pre post treatment data
				subid.base = mta[mta$trt==trtFlag & mta$spct==spcFlag & mta$time %in% b.times & !mta$samid %in% outliers,'subid' ];
				subid.post = mta[mta$trt==trtFlag & mta$spct==spcFlag & mta$time==time & !mta$samid %in% outliers,'subid'];
				subid.comp = intersect(subid.base,subid.post);
			
				## ensure there are data to compare
				if (length(subid.comp)!=0) {
				
					## get ids for treatment time
					id.time = mta[mta$trt==trtFlag & mta$time %in% c(b.times,time) & mta$subid %in% subid.comp & !mta$samid %in% outliers,'samid'];
			   
					## read gene sets to retain
					in.file.gene.sets = paste(in.dir.lcpm,'/',spcFlag,'_posttp_analysis_gene_set.tab.gz',sep='');
					gen.lst = read.table(gzfile(in.file.gene.sets),header=F,stringsAsFactors=F);
					
					###########################################
					##
					## FILTER EDGE_R OBJECT FOR POST treatment TIME 
					## 
					###########################################
					
					## retain columns in the count matrix that match selected libraries
					id.time.dge = match(intersect(id.time,colnames(dge.lab.time$counts)),colnames(dge.lab.time$counts));
					dge.lab.time$counts  = dge.lab.time$counts[,id.time.dge];
					
					## retain rows in the count matrix that match genes that passed the low expression cut off
					dge.lab.time$counts  = dge.lab.time$counts[match(gen.lst[,1],rownames(dge.lab.time$counts)),];
					
					## update sample matrix that match selected libraries
					dge.lab.time$samples = dge.lab.time$samples[id.time.dge,]
					
					## align meta data
					mta.spc.trt.time = mta[match(colnames(dge.lab.time$counts),mta$samid),];
						
					###########################################
					##
					## EXPERIMENTAL DESIGN SETUP
					##
					###########################################
					
					## set all treatment time points to 0;
					mta.spc.trt.time[mta.spc.trt.time$time %in% b.times,'time']=0;		
					
					## define factors and reference levels
					timef = factor(mta.spc.trt.time$time);
					subf = factor(mta.spc.trt.time$subid);
					
					## design matrix used for glm
					if(glm.model.paired==1) {
						design = model.matrix(~0+subf+timef);
					} else {
						design = model.matrix(~0+timef);
					}
					
					## select coefficient of interest (time x treatment interaction)
					coefPos = grep(paste('timef',time,sep=''),colnames(design))
					
					###########################################
					##
					## ESTIMATE DISPERSION
					##
					###########################################
					
					dge.lab.time=estimateGLMCommonDisp(dge.lab.time,design)
					dge.lab.time=estimateGLMTrendedDisp(dge.lab.time,design)
					dge.lab.time=estimateGLMTagwiseDisp(dge.lab.time,design)	
				
					###########################################
					##
					## FIT MODEL / LIKELIHOOD RATIO TEST / FILTERING
					##
					###########################################
					
					dge.lab.time.fit = glmFit(dge.lab.time,design);
					
					## execute likelihood ratio test 			
					dge.lab.time.lrt = glmLRT(dge.lab.time.fit,coef=coefPos);
					
					## get fold change and adjusted p_values
					dge.lab.time.all = topTags(dge.lab.time.lrt,adjust.method="BH", sort.by="logFC",n=50000);
					
					## filter gene lists for fold change and adjusted p_value
					dge.lab.time.sig = dge.lab.time.all$table[abs(dge.lab.time.all$table$logFC)>=log2(as.numeric(glm.sdeg.fold)) &
							dge.lab.time.all$table$FDR<glm.sdeg.qval,];	
							
					###########################################
					##
					## SAVE GLM RESULTS
					##
					###########################################
					
					## save R objects 	
					out.file.glm = paste(out.dir.glm,'/',spcFlag,'_',trtFlag,'_tp',time,'_glm.RData',sep='')
					glm = list();
					glm$DGEList = dge.lab.time;
					glm$DGEGLM  = dge.lab.time.fit;
					glm$DGELRT  = dge.lab.time.lrt;				
					save(glm,file=out.file.glm,compress=T);
					
					## save all tabular results	
					out.file.lab.time.all= paste(out.dir.glm,'/',spcFlag,'_',trtFlag,'_tp',time,'_glm_all.tab',sep='')
					write.table(dge.lab.time.all,file=out.file.lab.time.all,sep='\t',quote=F,row.names=T);
					R.utils::gzip(out.file.lab.time.all,overwrite=TRUE)
					
					## save significant tabular results	if there are anyu
					if(nrow(dge.lab.time.sig)>0) {
						out.file.lab.time.sig = paste(out.dir.glm,'/',spcFlag,'_',trtFlag,'_tp',time,'_glm_sig.tab',sep='')
						write.table(dge.lab.time.sig,file=out.file.lab.time.sig,sep='\t',quote=F,row.names=T);
						R.utils::gzip(out.file.lab.time.sig,overwrite=TRUE);
					}
				}
			}
		}
	}
}

Try changing these to seq_along(VAR)

It might not be the solution, but is less likely to throw problems in general, according to Reliable Sources.

Is it possible that your splits create some zero-length cases?

1 Like

Should I change all of length parts or just length (splits) part

I'd do all of 'em. It's good practice and may as well do it at one pass. Let everyone know if it makes a difference?

1 Like

So it becomes for example 1:seq_along(splits), seq_long(spcFlags) ? Or 1: length (splits) = seq_along(splits)

Just on the first line in the foreach call. You need to use the [] operator to subset.

This is what I did

foreach (z=seq_long(splits)).....

No, that shouldn’t work. That’s assigning and reassigning the successive values. Only use it inside forerach()

Still produced the error:

Can you post a reprex with representative data? See the FAQ: How to do a minimal reproducible example reprex for beginners. It doesn't have to be all the data or even your data, just so long as it acts the same way.

This topic was automatically closed 42 days after the last reply. New replies are no longer allowed.

If you have a query related to it or one of the replies, start a new topic and refer back with a link.