picard: Upstream deletions and CollectVariantCallingMetrics do not play nice right now.

The current VCF spec allows for a * allele (no brackets):

“The ‘*’ allele is reserved to indicate that the allele is missing due to a upstream deletion.”

CollectVariantCallingMetrics treats this as a third (size 1!!) allele so that in the case of

1   10347   .   TAAACCCTA   T   100 .   AC=2    GT  0/1 0/1
1   10350   .   A           C,* 100 .   AC=3    GT  1/2 0/2

both the 0/2 and 1/2 genotypes in the second line are counted towards TOTAL_MULTIALLELIC_SNPS (for the detailed metrics) Also, both of these genotype will not be counted towards the TOTAL_SNPS (as that only captures bi-alleleic SNPs). So upstream deletions are “hurting” both the monomorphic samples (as they get an inflated TOTAL_MULTIALLELIC_SNPS ) and the polymorphic samples (as they get a deflated TOTAL_SNPS count)

I propose changing this behavior so that an upstream deletion will count as the reference allele for the purpose of metrics.

I will also add a few column or two to capture the number of upstream deletions, perhaps counting the 0/2 separately from the 1/2 genotypes.

Does this sounds reasonable to folks?

@eitanbanks @tfenne ?