Fig. 5: The COSMIS score is strongly correlated with both pathogenicity and gene constraint level.

a COSMIS score distributions for 19,346 benign, 14,824 pathogenic, and 115,172 VUS sites from ClinVar (Methods). Pathogenic variants have significantly more constrained 3D spatial neighborhoods (COSMIS score median −1.1) than benign variants (median score 0.0) (\(p \, < \, 2.2\times {10}^{-308}\), two-sided Mann–Whitney U test). In boxplot graphs center line indicates median, bounds of box indicate 25th and 75th percentiles, and whiskers indicate minimum and maximum. b Odds ratio of ClinVar pathogenic variants versus benign variants for different COSMIS score percentile bins (lower bins correspond to more constrained COSMIS scores). Amino acid sites with lower COSMIS scores are enriched for pathogenic variants whereas sites with higher scores are depleted of pathogenic variants. Error bars indicate 95% confidence intervals. The horizontal dashed line indicates OR = 1. The values for each cell of the contingency table used for the OR calculation in each percentile bin were reported in Supplementary Data 14. c COSMIS score distributions of amino acid sites in six groups of proteins encoded by genes with different functional annotations (and the dataset as a whole). As the anticipated functional constraint on each category increases (top-to-bottom), amino acid sites in proteins in the category have more constrained COSMIS scores on average. In boxplot graphs center line indicates median, bounds of box indicate 25th and 75th percentiles, and whiskers indicate minimum and maximum. OR odds ratio, VUS variants of uncertain significance. Source data are provided as a Source Data file.