298T Poster - Population Genetics
Thursday June 09, 8:30 PM - 9:15 PM

Authors:
Chris Smith; Peter Ralph; Andrew Kern

Affiliation: University of Oregon

Keywords:
Theory & Method Development

Most organisms disperse a limited distance from their birth location. This limited dispersal shapes patterns of genetic variation over a landscape, such that individuals that are close to one another in space will be more closely related than those that are spatially distant. This pattern, sometimes called isolation by distance, in turn can be used to infer spatial population genetic parameters. Here we demonstrate how deep neural networks can be used in combination with geographically-referenced genotype data to estimate a critical spatial population genetic parameter, σ, the mean per-generation dispersal distance. Using extensive simulation, we show that our deep learning approach is competitive with or outperforms state-of-the-art methods, particularly at small sample sizes (e.g., n=10). Whereas competing methods depend on accurate identification of identity-by-descent tracts or information about local population density as input, our method uses only single nucleotide variants and spatial coordinates as input. These features make our method, which we call dspp, a potentially valuable new tool for estimating dispersal in non-model systems. We demonstrate dspp on two such datasets with publicly available data, Anopheles gambiae and Arabidopsis lyrata. Finally, we consider the future utility of deep learning methods for spatial population genetic inference.