analysis of single and PE reads with htseq-count and DESeq

[dglemay]

Hello,

I've analyzed some single and paired-end reads that were aligned to the human genome using TopHat and then put them through the htseq-count to DESeq pipeline. A clustering analysis of the variance-stabilized transformed count data (thank you Simon!) suggests that the differences between library types (single vs paired-end) are much bigger than the very significant differences between biological conditions.

Why might the count data between single and paired-end library types be so different?

Thanks,
Danielle

[Simon Anders]

If you want to get to the bottom of this, you could, for example, redo the alignment and counting of your paired-end data but using only the fastq files from the first pass (the first end). After all, library-prep is the same for single-end and paired-end, and hence, if you simply throw away the second end, the difference should vanish. If so, the whole issue is an artifact of alignment or counting algorithms having subtly different biases when working with paired-end data. If not, it might be some other batch effect that only happens to be confounded with library type.

Also note the example in the DESeq vignette that shows how to account for such differences by introducing a blocking factor for library type. This only works if your library type is not confounded with your treatment, of course.