Inference of Gene-Level Selection in Human Populations: New Methods and Applications

Recent large-scale exome and whole genome sequencing studies enable a systematic view of the allelic spectrum in humans. This data can be used to infer the selection forces acting on human populations, but new methods need to be developed for this task, mainly due to the need to aggregate signal from multiple variants.
I will present a model and method for estimating gene-specific selection parameters from intra-human variation present in large-scale exome sequencing data in multiple human populations.
I formulate a generative probabilistic model for the data, based on an extension of the classic Wright-Fisher process, and use it to estimate demographic and selection parameters. The model integrates selection signals from different types of variants in the same gene (e.g. missense and loss-of-function variants) into a single framework.

I will also present an analysis of the gnomad dataset containing ~120,000 different exomes from 11 human populations, utilizing the developed framework for several applications: (i) increasing the accuracy of our selection coefficient estimators by aggregating information from multiple European populations (ii) testing for differences in selection between European, African and Asian populations (iii) improving statistical power in rare-variants association studies (iv) comparison of intra-species and inter-species selection estimates showing systematic changes in selection coefficients during human history, and (v) comparison of gene sequence-based constraint vs. genomic structural constraint, measured by intolerance to copy-number variants, showing significant differences between these two types of constraint.