Dear Antti,

I want use your BitSeq package for transcriptome analysis and want to get transcript counts (for usage with other tools).
If I set "type=counts" in getExpression all transcripts have counts of 1 or more. I think this is that you can logarithmise the data afterwards. Is it safe to substract 1 of this counts to get real counts (i.e. are transcripts with 2 counts initially called with one real count and 1 is added, or are only all zeros replaced by 1)?

Thank you,
Dietmar

Dear Dietmar,

Thanks a lot for your question!

The minimum count 1 comes from the prior of our Bayesian model, which effectively adds 1 "pseudocount" for each transcript. The expected number of reads in the sample aligning to a transcript with reported count c is thus indeed c-1.

For further processing, I would recommend first filtering out at least all transcripts with average reported count smaller than 1.1, as they are very likely to be unexpressed. What to do with other counts may depend on what is the next tool and what are its assumptions, but subtracting the extra pseudocount to get raw read counts for tools that expect those sounds at least like a reasonable starting point.

Antti