Identifying differential selection between humans and primates through population modeling

We first established a neutral background distribution of mutation rates per gene for each primate species by fitting the Poisson Random Field model to the segregating synonymous variants in each species. The observed number of segregating synonymous sites is a Poisson random variable, with the mean determined by mutation rate, demography, and sample size (34). For simplicity, we assumed an equilibrium (i.e., constant) demography for all species besides humans; for humans, we used Moments (51) to find a best-fitting demographic history based on the folded site frequency spectrum of synonymous sites. We adopted a Gamma distributed prior on mutation rates, which also accounts for the impact of GC content on mutation rate. We optimized the prior parameters through maximum likelihood and computed the posterior distribution of the mutation rate per gene.

The number of segregating nonsynonymous sites is modeled as a Poisson random variable similar to synonymous sites with additional selection parameters. We assumed that every nonsynonymous mutation in a gene shares the same population-scaled selection coefficient γ_ig. To explicitly estimate the selection coefficient of each gene per species, we devised a two-step procedure analogous to an expectation-maximization algorithm to control for differences in population size across species.

To identify genes in which human constraint is different from nonhuman primate selection, we developed a likelihood ratio test to test whether population-scaled selection coefficients are significantly different between humans and other primates. We then assessed whether our population genetic modeling improved the correlation of selection estimates of our primate data with previous gene-constraint metrics in humans, including pLI (28) and s_het (111). To validate the performance of our model, we performed population genetic simulations.