Good evening,
I am doing a research on 30 samples divided by condition. I performed first a pipeline with salmon alignments and DESeq2, and then another pipeline with STAR+ featureCounts and edgeR for the DGE. I am encountering several differences in between. The most important is that the lib.size of one sample processed in the pipeline STAR+featureCounts (say it's P10) is very low with respect to other samples. Specifically, it differ to some order of magnitude. Thing that does not happen with the sampled processed with salmon. I am sharing the screen of the libsize of, respectively, sampled processed with salmon and samples processed with STAR+featureCounts.
I checked the percentage of successfully done assignments with featureCounts and, for P10, it's the same as others. Seems that nothing differ. STAR percentage also is good: about 91%. Also tried to re-align everything, but the same problem persists. I also share the R script executed to do the DGE with the STAR+featureCounts counts.
library(edgeR)
d <- DGEList(counts=df,group=sampleTable$condition)
# filtering data
keep <- rowSums(cpm(d)>100) >= 2
d <- d[keep,]
dim(d)
d$samples$lib.size <- colSums(d$counts)
d$samples
d <- calcNormFactors(d)
d
... continue with glmFit and plots
