Evaluation regions for HG002 about deepvariant HOT 3 CLOSED

Thank you! We weren't 100% certain the ideal dataset to evaluate the callset on. Personally I like to use two evaluation intervals: RefSeq alone and the capture intervals. RefSeq is nice because it is entirely orthogonal to the capture approach, which I feel is fair since the capture intervals are often only those where the capture proves to be effective, not actually where one wants to analyze their exome data (closer to RefSeq IMO). But I like replacing our refseq + 50 bp intervals with the capture intervals.

Out of curiosity, why intersect the capture intervals upfront with the confident regions? Happy and VCFEval both will internally do that intersection if you call on the capture intervals and provide the confident regions for evaluation?

from deepvariant.

That's a good question. We use this BED file to provide to the calling programs where possible so that we don't use compute performing calling in regions we won't be evaluating on. You are correct, doing the intersection in hap.py will work just fine in terms of the final results.

Here is the non-intersected BED file:

*edited by adding file attachment
agilent_sureselect_human_all_exon_v5_b37_targets.bed.gz

from deepvariant.

Thanks Andrew! We've put this on our internal buganizer component for DeepVariant and we'll update the case study to use this.

from deepvariant.