Currently QCTOOL outputs 'GP' field in VCF files but the contents are unscaled probabilities. However, the VCF format specified these should be PHRED-scaled probabilities. Here is one way of going about it suggested by Warren Kretzschmar:
- switch to using phred scaled probabilities in the next release
- use a command line switch for legacy support of unscaled genotype probabilities for input and output
- test all input genotype probabilities to see if they look like they are unscaled and die with the requirement for a "force phred scaled input genotype probabilities" command line switch if the input looks too much like unscaled genotype probabilities.
Some tests that might be useful: - do the probabilities sum very close to 1? - are any of the probabilities substantially greater than 1?
Also, I think the GP field is not documented in the qctool documentation. It would be good to document how the GP field is interpreted in any case.