Identifying complex modes of adaptation from population-genomic data

Project Summary Low-cost DNA sequencing has provided researchers with abundant genomic data in which to search for the unique footprints left by natural selection. However, many nonadaptive forces can obscure these signals, making it important to develop statistical methods that can account for multiple factors that influence genetic variation. My research in this area focuses on the design and application of statistical approaches for identifying regions undergoing positive selection, which increases the frequency of beneficial alleles in a population, and balancing selection, which maintains the frequency of distinct alleles in a population. During the past four years of my MIRA ESI, my group contributed to a number of methodological advances, including designing the first likelihood methods to detect positive selection from haplotype distributions, state-of-the-art likelihood methods to detect ancient balancing selection within and across species, the first likelihood method to detect positive selection by adaptive introgression, and the first machine learning method to detect positive selection by explicitly modeling genomic autocorrelation. Applications of our methods to empirical data led to several novel insights, including evidence of convergent positive selection in Europeans and East Asians, positive selection at olfactory genes that affect communication and behavior in New York City rats, and balancing selection at venom genes that may influence predator-prey coevolutionary dynamics in rattlesnakes. During the next five years, I propose to develop improved statistical and machine learning approaches to detect complex modes of adaptation by leveraging information about how different evolutionary forces simultaneously shape the distribution of genetic diversity around adaptive sites. In particular, my future research program will be subdivided into several interrelated goals: designing likelihood frameworks to identify positive and balancing selection while accounting for genomic, temporal, and spatial autocorrelations; developing methods to identify regions that underwent complex positive and balancing selection from ancient variation; introducing signal processing approaches to extract features from images of genomic variation for use in machine learning models of adaptation; and building innovative domain adaptation procedures to circumvent genetic and demographic parameter uncertainty when training machine learning predictors of adaptation. Availability of these methods will empower researchers to address important questions about adaptation in model and non-model organisms from ancient and contemporary samples, increasing inclusivity in the field, and thereby broadening our knowledge about adaptation. Further, we will apply these methods to answer or revisit key evolutionary questions about the roles of different adaptive mechanisms in primates, rodents, snakes, insects, plants, and other organisms with complex or unknown demographic histories. Advantages of these studies are two-fold, in that they will yield powerful new approaches for identifying signatures of diverse modes of adaptation from genomic data, as well as elucidate evolutionary forces underlying the acquisition of adaptive phenotypes, such as those involved in disease resistance and pathogen defense.