Effects of applying BQSR twice?

Question

[Chris Miller]

Does anyone have insight into what happens if you apply BQSR twice to a sample?

The case here is that we might run a pipeline that aligns and runs BQSR, then we throw away the original fastqs. If we later want to realign, we can regenerate fastqs, but the original quality scores are gone. What would the effect be of running them through a pipeline that applies BQSR a second time?

Obviously we could modify the pipeline to omit BQSR on a subsequent run, but I wonder if that's even necessary. Anyone with more insight into the underlying algorithm know what the effect would be? In general terms, is it going to be worse, or largely inconsequential?

Answer 1

[h.mon]

I suspect if same reference genome and known variants are used for both rounds of BQSR, the second BQSR would produce little change over the first round. If different versions of reference genome and known variants are used, there may be additional changes, but honestly it is just a guess.

This sounds like a cool little project for a quarantine, no? One could either use simulated reads, or "gold standard" data sets, to investigate the issue.

[Chris Miller]

Sure - straightforward enough to check, but i was hoping someone else had already done the comparison :)