B. Variant Predictions

The tolerated or damaging effect of each variant in the dataset was obtained from the following widely used variant predictors: SIFT [2], PolyPhen2 [3], MutationAssessor [11], and Condel [12]. SIFT and MutationAssessor assess the effect of a variant based on its sequence conservation, whereas PolyPhen2 uses sequence conservation and structural features. Condel integrates predictions from two tools, MutationAssessor and FATHMM [13], to generate a new consensus score [12]. The default threshold recommended for each program was used to predict the damaging or tolerated effect of each variant. We applied a binary definition to each variant: damaging vs tolerated. Because PolyPhen2 and MutationAssessor provide multiple effect definitions, we considered the PolyPhen2 predictions “probable deleterious” and “possibly deleterious” and the MutationAssessor predictions “high” and “medium” impact [11] as “damaging,” whereas the PolyPhen2 score “benign” and the MutationAssessor predictions “neutral” and “low” impact as “tolerated.”