Creating gene counts for differential gene expression analysis

Question

[roger.huerlimann]

Hi all,

I was wondering if there are any pros and cons for using a genome guided versus pseudo-mapping approach for differential gene expression analysis.

I have a decent genome assembly with associated gene models. I was using STAR to get gene counts for my RNA-seq data using the genome and gtf file. However, I've seen people recommend using the transcriptome instead with a pseudo-mapper like salmon or kallisto. I know pseudo-mappers are faster, but is there any other benefit of going down that route? The last option would be to use STAR with RSEM, but I'm not sure if that gives me any benefits over just using STAR with inbuilt HTSeq counting.

Cheers, Roger

Answer 1

[swbarnes2]

RSEM should be smarter than STAR's built-in htseq-count. It handles ambiguous reads better.

[dsull]

Agreed.

Also a benefit to kallisto/salmon over STAR+RSEM aside from speed/memory is that you can estimate inferential uncertainty which improves your differential expression analysis.

[roger.huerlimann]

Thank you both for your replies!

Estimating the inferential uncertainty sounds interesting, I will have to look into that.