Figure 2: Factors that contribute to our estimates of $s_{\text{het}}$.

A) Likelihood curves for different allele frequencies (f) and mutation rates. B) Scatterplot of $s_{\text{het}}$ estimated from LOF data (y-axis; posterior mean from a model without features) against the prior’s predictions of $s_{\text{het}}$ (x-axis; mean of learned prior). Dotted line denotes $y=x$. Each point is a gene, colored by the expected number of LOFs. C) Comparison of posterior distributions of $s_{\text{het}}$ (95% Credible Intervals) from a model with (blue lines) and without (orange lines) gene features. Genes are ordered by their posterior mean in the model with gene features. D) Top: scatterplot of LOEUF (y-axis) and our $s_{\text{het}}$ estimates (x-axis; posterior mean). Each point is a gene, colored by the expected number of LOFs. Bottom: scatterplot of $s_{\text{het}}$ estimates from [4] (y-axis; posterior mode) and our $s_{\text{het}}$ estimates (x-axis; posterior mean). Numbered points refer to genes in panels E and F. E) TBC1D3 and PLN are two example genes where the gene features substantially affect the posterior. We plot their posterior distributions (blue) and likelihoods (orange; rescaled so that the area under the curve = 1). F) AARD and TWIST1 are two example genes with the same LOEUF but different $s_{\text{het}}$. Posteriors and likelihoods are plotted as in panel E.