Dealing with multimapping reads in featureCounts

Question

[Sreeraj Thamban]

Hi Biostars, I used featureCounts to generate the counts table for the DEG analysis of my RNASeq data. I didn't count multi-mapping reads, but one of my libraries has 33% multi-mapped reads. I am afraid if I exclude the multi-mapping reads I will end up loosing a significant portion of the information. Please suggest me whether I should include or exclude multi-mapping reads in featureCounts. Whether these settings will make significant changes in the DEG analysis result?

[grant.hovhannisyan]

you might try to tune parameters of aligner (like maximum number of mismatches) to reduce the multimapping rate.

Answer 1

[bruce.moran]

The only way to know if multi-mapping reads make a difference to a DE analysis is to run it with and without them. In my experience, it depends on the study. Cell lines seem more robust, tissue samples less, but this may be due to inherent biological variation. There are several ways to use mulit-mapping reads, see this review for overview. A new method is mmquant which will replace featureCounts in your pipeline, or there is a new gene-level aggregating method which you will have to use on transcript-level counts.

Good luck, please post back if you find something of interest.

[Sreeraj Thamban]

Hello Bruce, I did the DE analysis with and without multi-mapping reads and the results were almost similar. Did you use mmquant? I read the paper and planning to give it a try in the future.

[Taha]

Hi,

I am facing a similar issue, if you tried mmquant, has the utilization of mmquant been beneficial?