I am analyzing tumor DNA sequencing data to perform variant calling. I need to find the minimum number of DNA strands (sample size; independent samples, 'depth of coverage') needed to detect mutations occurring at frequencies of 2-5%, with 95% confidence, given a 1% background mutation rate, at power levels of 0.8, 0.9, and 0.99.
Past collaborators did this, using a "cloglog Binomial distribution"(?), and got some of the following results:
frequency: 0.02 power: 0.8 alpha: 0.05 sample size: 1239 frequency: 0.03 power: 0.8 alpha: 0.05 sample size: 423 frequency: 0.04 power: 0.90 alpha: 0.05 sample size: 299 frequency: 0.05 power: 0.99 alpha: 0.05 sample size: 315
I am trying to replicate their analysis in R, to validate and fill in more values, but it seems like I am doing something wrong because I am not getting the same values. Using the
library("pwr") pwr.p.test(h = 0.02, sig.level = 0.05, power = 0.80, alternative = "greater")
proportion power calculation for binomial distribution (arcsine transformation) h = 0.02 n = 15456.39 sig.level = 0.05 power = 0.8 alternative = greater
Here, it is giving me an
n of 15456, when the value should be 1239
As per the docs for this package, the 'effect size' is important, so I am wondering if that might be the source of the discrepancy? And I am not sure how the 'coglog Binomial distribution' plays into it, especially since
pwr says it uses an 'arcsine transformation' instead.