featureCounts parameters

Question

[Lucy]

Hi,

I was wondering whether you would recommend adjusting the following featureCounts parameters and if so, why?

--fracOverlap, --fracOverlapFeature, --minOverlap, --maxMOp

I am trying to decide which parameters to go for. I have high quality bulk RNA-seq data (75 bp or 150 bp paired-end reads) from humans and I have mapped the data with STAR. Please let me know if you require any other information.

Best wishes,

Lucy

[Shred]

Are you doing a Differential Expression analysis? If so, are you interested in a Gene Differential Expression or Transcript Differential Expression?

[Shred]

Please use comments to reply, instead of answers. Other info:

• How's your read length distribution?
• Which is the experimental design? Is there something which may benefit from accounting multimapping reads (considering a PE experiment)?
• Tools like FeatureCounts or HtSeq were replaced for more accurate ones like Kallisto or Salmon. Why not using them?

[Lucy]

Hi Shred,

I haven't actually seen any convincing evidence that Kallisto and Salmon are more accurate than tools like featureCounts. Please supply the references.

My read lengths are 75 bp or 150 bp paired-end (and I have done a small amount of trimming). I will run featureCounts twice, once excluding multi-mapping reads and once counting them fractionally.

Best wishes,

Lucy

[Istvan Albert]

Salmon is better suited for transcript-level redistribution of reads. Featurecounts was not designed to distinguish between transcripts.

Note how Salmon is not a replacement of Featurecounts but a replacement of an entire two-step process: alignment+counting

When it comes to finding coverages at the gene level, the differences between the classification methods of Salmon vs the alignments +counting are probably less pronounced. Though running Salmon is probably much faster and simpler.

The main downside of Salmon is that you are not getting a genome level alignment that you could visualize, thus I always recommend doing both types of analyses and comparing the results.

[Shred]

Salmon could be used to quantify .bam files aligned against the transcriptome fasta file, which, for the purpose of a DE analysis with human RNAseq data, may be enough.

[Shred]

Here you go https://genomebiology.biomedcentral.com/articles/10.1186/s13059-018-1419-z

[Lucy]

Yes I will be doing gene level differential expression analysis.