Differential Expression Analysis with Salmon vs Genomic Aligners Like Star

Question

[dk0319]

Can anyone with experience using aligners like Salmon and STAR for RNA-seq comment on the strengths and weakness of one method over another. So far I have performed the alignments and quantification and noticed some pronounced differences in the generated counts and I was curious to hear the thoughts of more experienced users.

[GenoMax]

STAR and salmon are using two methodologies. While STAR is aligning the reads, salmon is using quasi-mapping. You can read more about the differences from @Rob Patro (author of salmon) here: C: Mapping vs Quasi-Mapping

[dk0319]

I ended up finding this, which offers a nice overview of the differences (A: Could you explain the difference between STAR, KALLISTO, SALMON etc. to experime ). I am still curious to hear from people with experience working with both approaches to see if they ever encountered any issues.

Answer 1

[swbarnes2]

How are you generating read counts with STAR? Its built-in algorithm is not as smart as Salmon.

[dk0319]

I used the internal quantmode along with ht-seq and feature counts

[swbarnes2]

The internal mode should be the same as htseq-count. But what I said still stands; Salmon will be smarter than htseq or featureCounts at dealing with ambiguous reads.

Whether STAR + RSEM is significantly better than Salmon, that's a different question, because RSEM is smart about handling ambiguous reads

[dsull]

Correct; kallisto and salmon quantification are much better. See:

https://twitter.com/lpachter/status/1060597618479267840