When we run HaplotypeCaller, we often miss out on variants due to the assembly graph only being able to find the reference haplotype (especially when there are repeats in the reference such that we have to use a really big kmer size to try to find variant haplotypes). It would be nice if there was an argument we had that would enable us to use UnifiedGenotyper to find variants in these cases. Unfortunately, the methods we would need from UnifiedGenotyper haven't been ported over to GATK4 (I think we would need calculateLikelihoodsAndGenotypes) in order to improve our sensitivity and capture sites where our assembly graph can't in HaplotypeCaller.

(This is from work done for https://github.com/broadinstitute/palantir/issues/506)