featureCounts vs transcript-aware quantification (Kallisto/Salmon)
Hello all,
I suppose I am musing a bit and wanted to discuss with other bioinformaticians. I am a head bioinformatician in my academic department. A few months ago, I was given new bulk RNA-Seq data to analyze alongside older data that was already part of a peer-reviewed manuscript (that I was not part of). I used a STAR --> Salmon alignment-based quantification method. After sending the DE analysis and "raw" expression values for all genes, I received word that my Salmon results for the published data and the original data differed greatly. The older data was processed via featureCounts, which is known to undercount genes with multiple isoforms. I spent a few weeks working backwards to determine what parameters were used in the published manuscript, and I confirmed that the "gold standard" featureCounts parameter set was used, which definitionally excludes any read that overlaps multiple "features", or is ambiguous between isoforms of the same gene. To resolve this, you would use the -O flag, etc etc.
I guess my complaint is, how is this acceptable? How can a very popular and widely-used program such as featureCounts exclude reads that overlap the same exon (that resides in different isoforms) by default? This default method is undercounting genes with multiple isoforms, and I see discussion of this exact issue online since 2015. Discussion of this issue has also been published.
To be brief, I am mainly concerned that a widely-used tool is undercounting isoform-laden genes by default and causing consternation for groups who don't have trained bioinformaticians on their team who have the time to look into these issues.
Thank you for listening to my rant, haha.