Provide reference distributions of evidence metrics and random forest cutoffs #880
Description
Module(s) or script(s) involved
- GenerateBatchMetrics
- FilterBatchSites
Description
We've observed multiple instances callsets where the distribution of an evidence metric used to train the Random Forest model can be skewed across the entire cohort. This has led to cutoffs deviating far from their usual values, which in turn can lead to problems during genotyping - over-genotyping of variants of a particular type, under-genotyping to miss a specific subset of variants, etc.
As a result, it would be useful to provide reference distributions of these metrics, as well as the cutoffs that they lead to. These reference distributions can come from known cohorts (e.g. gnomAD, 1KGP, etc) to ensure validity of our own distributions.