Extended Data Fig. 7. Genebass summary statistics.
(a) Gene-phenotype association strength (-log10 p-value) for each model, assessed across varying single-variant p-value thresholds from Genebass summary statistics. (b) Association strength across different SKAT-O missense p-value thresholds used to define gene-phenotype pairs. VESM + + and VESM-3B consistently outperform alternative models across thresholds. (c) Average Pearson correlation between predicted variant scores and Genebass β coefficients across 103 high-confidence gene-phenotype pairs. (d) Average association strength (-log10 regression p-value; two-sided test; scipy.stats.pearsonr) of all VESM and baseline models across phenotype categories, including gene-phenotype pairs outside the three main categories shown in Fig. 6b. (e) Correlation between VESM scores and Genebass effect sizes for gene-phenotype pairs not included in the main phenotype categories (complementing Fig. 6c).
