Effect size in power analysis when dealing with Poisson based variant caller

Question

[CY]

I was trying to estimate the minimum sequencing depth required to confidently detect somatic variant using Poisson distribution based variant caller (such as somatic variant caller or Pisces from illumina).

Power analysis is used which requires significance level (0.05), power / sensitivity (99%) and effect size to estimate sample size (sequencing depth). I am not sure how to calculate the effect size in such context. Can anyone share some insight? Thanks a lot.

[Kevin Blighe]

Yeh, but, a somatic mutation / variant at which frequency? Technically speaking, in a bulk biopsy of 1 million cells, a rare sub-population of 100 cells may harbour the key mutation that will eventually bring about metastasis. The frequency of this would be ~0.01%, though. You may want to simulate a pileup that follows a Poisson (based on its read depth) and 'randomly' impute this with variants at 'random' frequencies. Then, you could go about calculating what you need. Just typing out loud.

[CY]

I will have to give up detecting somatic variant with very low frequency (<1%) since matched normal is absence. In terms of intuitive understanding of effect size of somatic variant. I guess the VAF can be viewed as effect size(say, the frequency of noise is 1%, higher VAF of variant indicates greater effect size of such somatic variant). Does this round reasonable?

Answer 1

[markus.riester]

Such a power calculation is described in Carter el al. You can find a well documented implementation in our PureCN R package.

Essentially this calculates the probability of sampling sufficient supporting reads so that you can rule out a sequencing error.