Mutect isActive loses sensitivity when allele fraction is comparable to base error rate

[davidbenjamin]

If you look at the approximation going from equation 34 to 35 in https://github.com/broadinstitute/gatk/blob/master/docs/mutect/mutect.pdf you will find that we replace f(1 - e) + (1 - f)e by just f(1 - e), where f is the allele fraction and e is the error rate. When f is much bigger than e this is okay but when they are comparable (consider mitochondrial or cfDNA calling with f = 1% and base qualities of 25) the approximation breaks down and we significantly underestimate the log odds, thereby failing to consider a region active.

This must be fixed!

@meganshand

[meganshand]

@fleharty This might be of interest.

[davidbenjamin]

Closed by #4832.