Distribution underlying VCF QUAL - all sites or just those in VCF?

Question

[Jeff.B.Gaither]

I'm wondering about the distribution which underlies a VCF QUAL score. Specifically, I understand that

"A site with QUAL=20 has a 99% chance of being a true mutation," or more precisely (I think)

"The collection of sites with QUAL=20 has the property that 99% of it consists of true mutations."

My question is, does "the collection of sites with QUAL=20" apply to just the sites listed in the VCF, or to all the sites sequenced in the experiment? This actually affects the definition of QUAL, I think, and if the latter it might mean we could expect millions of false positives in whole-genome variant-calls.

Thanks for any help! And sorry if I'm overlooking something, but I haven't found any documentation that addresses this point in a clear way.

Answer 1

[Jeff.B.Gaither]

Actually, perhaps a decent sample space would be, all the distinct calls made by the variant caller (in this case mpileup) by everyone in all of history. That is, 99% of the sites ever observed in history with QUAL=20 should be true mutations. A more formal definition would perhaps involve all possible return states of the variant caller...