How useful is ModelScore? #121
Description
Noticed some PMS2 variants we are interested in getting -2 in "Variant call quality filter" (+3 for . variant and -5 for low quality ModelScore).
-5 is a quite big penalty. This causes the variant to go from rank ~7 to ~34. The calls are indeed quite bad, but we are still interested in them in this case. I wonder how useful the ModelScore penalty is for DeepVariant/LRS, and how many variants we would add if we remove it.