Fig. 2: Genome-wide profile and performance of gwRVIS.

a Intolerance to variation profile across all genomic windows (having 3 kb length). A regression line (shown in green) is fit between “all” and “common” (MAF > 0.1%) variants across all windows. gwRVIS can be visualized as the vertical distance of each data point from the regression line (prior to normalization by the standard deviation of the total distribution). Red dots represent the top 1% of most intolerant windows (i.e., having fewer common variants than expected) while blue dots represent the top 1% of most tolerant windows. b gwRVIS distribution across all chromosomes, as extracted from the TOPMed dataset. A highly tolerant set of windows in chromosome 6 is enriched for HLA complex regions. c gwRVIS scores distribution (with single-nucleotide resolution) across different sets of mutually exclusive CCDS windows: OMIM-Haploinsufficient, 25% most intolerant (based on RVIS), 25% most tolerant, and rest of CCDS. P values from two-sided Mann–Whitney U tests are also provided for each pair of “adjacent” coding region classes in order of increasing intolerance. d Distribution of gwRVIS scores across different coding and noncoding genomic classes, in descending order of intolerance to variation: UCNEs, VISTA enhancers, UTRs, CCDS, introns, lincRNAs, and intergenic regions. The red horizontal dashed line (gwRVIS = 0) represents the mean of the theoretical null distribution (i.e., where the observed number of common variants equals the expected number). Intergenic regions are normally distributed around the null distribution, which validates their use as an empirical null distribution. Two-sided Mann–Whitney U has been employed to compare the gwRVIS distributions across all pairs of genomic classes (***p < 1 × 10⁻³⁰⁸). For each boxplot, its central line represents the median, the bounds represent the 25th and 75th percentile, and the whiskers extend up to 1.5 the interquartile range from the respective bounds.