Is it possible to get a description of how the gnomAD files (for Mutect2 and CalculateContamination) were generated? Additionally is possible to get some guideline on how they should be generated for non-human species (since gnomAD is human only)?
I saw the wdl script on GitHub. It's very helpful, but incomplete. For example, there is a "minimum_allele_frequency" variable that is expected as an input, so its value is unknown. Also, there is some regex that expects a specifically formatted VCF. Since gnomAD provides only chromosome VCFs, it's possible that the whole genome VCF would differ depending on how exactly those are merged. The VCF can be acquired from other sources (such as Ensembl) and that will also have slightly different fields.
More generally, af-only-gnomad.hg38.vcf.gz contains ~260M records, but small_exac_common_3.hg38.vcf.gz has only ~62K. That is very heavily filtered. That seems like extremely heavy filtering.