We could save a lot of time in PairHMM if we could identify reads supporting the reference allele and run PairHMM only on reads supporting alt alleles and a random sample of ref reads. This could especially save a lot of time on deep, low-AF sequencing, such as cfDNA. One simple criterion for ref-ness that might work would be whether a read gets threaded onto the reference path in the assembly graph.
We could save a lot of time in PairHMM if we could identify reads supporting the reference allele and run PairHMM only on reads supporting alt alleles and a random sample of ref reads. This could especially save a lot of time on deep, low-AF sequencing, such as cfDNA. One simple criterion for ref-ness that might work would be whether a read gets threaded onto the reference path in the assembly graph.