What determines an active region? Different HaplotypeCaller results for small vs. large cohorts
We observed a de-novo mutation identified by HC in a single trio that was not called when the same trio is included in a larger 95-sample cohort. The immediate cause seems to be that in the larger cohort the region surrounding this variant is no longer considered active. I also observe that whether the region is "active" depends on how large of an interval I supply. A 100bp interval centered on the site has active regions, a 300bp interval does not. When I "forceActive" the variant is called as expected in all intervals.
I can't find much information about how active regions are determined. At first, I hypothesized that the activity profile scores were smaller in the larger cohort (and thus no longer crossing the fixed threshold in ActivitiProfile.java, but the scores as dumped by --activityProfileOut look to have similar magnitude. What other information does active region determination take into account? How can I interpret the --activeRegionOut output? For the 300bp interval where the variant (@ chr3:51992816) is not called I get
#track graphType=line Chromosome Start End Feature ActiveRegions chr3 51992665 51992666 end-marker 0.00000 chr3 51992665 51992778 size=113 -1.00000 chr3 51992778 51992779 end-marker 0.00000 chr3 51992778 51992948 size=170 1.00000 chr3 51992948 51992949 end-marker 0.00000 chr3 51992948 51992966 size=18 -1.00000
When I use a smaller 100bp interval, I get the following active regions output:
#track graphType=line Chromosome Start End Feature ActiveRegions chr3 51992765 51992766 end-marker 0.00000 chr3 51992765 51992778 size=13 -1.00000 chr3 51992778 51992779 end-marker 0.00000 chr3 51992778 51992853 size=75 1.00000 chr3 51992853 51992854 end-marker 0.00000 chr3 51992853 51992866 size=13 -1.00000
The one noticeable difference between the two is the length of the potential active regions (75 vs 170), does that impact anything? Why despite the above output suggesting that there is indeed an active region does local assembly not get performed? Of note, in the larger cohort there is another singleton variant 82bp downstream in another sample (i.e. not in the original trio). I presume it is that variant that is leading to the longer active region.