Ambiguous Reads Treatment by HTSeq-Count

[Fernas]

Currently, if a person uses HTSeq-Count to count reads on transcript level and the tool finds a read that overlaps two overlapped transcripts (this is very common) then the tool will identify the read as ambiguous and will not include it in the counting process. As the author assumed, this is useful for identification of differentially expressed genes. However, this approach will underestimate the expression of overlapping transcripts and overestimate unique ones.

Can we modify somehow HTSeq-Count to include such reads? any other tools that overcome such situation?

[dpryan]

htseq-count isn't intended for that use, so while you could modify it, why both? You can get featureCounts to do this, but you really need to be careful about what you then do with these counts. If you're going to use these in something like DESeq2 or edgeR then you're simply doing it wrong (you'll just simply get wrong results).

In general, transcript abundance is better estimated with an expectation maximization algorithm (e.g. FluxCapacitor, or eXpress).

[Fernas]

Thank you very much @dpryan for the very informative reply. I will have a look into these tools.