Predicting geographic location from genetic variation with deep neural networks

Abstract

Most organisms are more closely related to nearby than distant members of their species, creating spatial autocorrelations in genetic data. This allows us to predict the location of origin of a genetic sample by comparing it to a set of samples of known geographic origin. Here, we describe a deep learning method, which we call Locator, to accomplish this task faster and more accurately than existing approaches. In simulations, Locator infers sample location to within 4.1 generations of dispersal and runs at least an order of magnitude faster than a recent model-based approach. We leverage Locator’s computational efficiency to predict locations separately in windows across the genome, which allows us to both quantify uncertainty and describe the mosaic ancestry and patterns of geographic mixing that characterize many populations. Applied to whole-genome sequence data from Plasmodium parasites, Anopheles mosquitoes, and global human populations, this approach yields median test errors of 16.9km, 5.7km, and 85km, respectively.

Spatial population genomics of a recent mosquito invasion

“…Despite high regional gene flow, we managed to locate the source of 2-3 of the 4 incursions below the cordon sanitaire. Considering the absence of discrete populations within the TSI, the genotype-based method of Locator (Battey et al, 2020a) was better suited to this dataset than the allele-frequency based method of assignPOP (Chen et al, 2018), though the latter still performed well for Inc-1 and Inc-2 and both methods successfully traced samples collected 10 months after the initial collections. Although Locator may best be suited to non-clustered sampling designs that are rare in island systems, Locator may sidestep issues relating to invasion age and frequent gene flow by not requiring that reference genotypes be sorted into populations.…”

“…To estimate source locations of the four incursive Ae. albopictus, we used two complementary methods in the programs assignPOP (Chen et al, 2018) and Locator (Battey et al, 2020a). assignPOP treats each village as a population, and generates assignment probabilities to each hypothetical population using a Monte Carlo assignment test with a support vector machine predictive model.…”

“…albopictus to predefined populations is viable at broad scales (Schmidt, Chung, van Rooyen, et al, 2020), high regional coancestry in recent invasions may confound assignment. A new deep learning method, Locator (Battey, Ralph, & Kern, 2020a), infers origins from georeferenced genotypes directly and may be better suited for recent invasions.…”

“…While intragenerational dispersal has been investigated in ecological studies that observe the movement of individuals (Harrington et al, 2005;Schultz & Crone, 2001), it has up until recently been too expensive to accurately assess individual movement with molecular approaches as this requires many individuals to be sequenced at many loci. Current population genomic approaches such as those that investigate the spatial distribution of close kin (Escoda, Fernández-González, & Castresana, 2019;Escoda, González-Esteban, Gómez, & Castresana, 2017;Fountain et al, 2018;Jasper, Schmidt, Ahmad, Sinkins, & Hoffmann, 2019) or trace migrants to their specific origins (Battey, Ralph, & Kern, 2020;Schmidt et al, 2019) are helping build a framework for population genetics to detect and interpret the movement of individuals. This in turn is allowing studies to assess patterns of identity by descent across a range of temporal scales, from a single generation of movement to many generations in the past (Bradburd & Ralph, 2019).…”

Reconstructing the history and biological consequences of a plant invasion on the Galápagos islands

“…In general, pairwise sequence divergence was lower in Galápagos-Ecuador comparisons (average dxy = 2.3 x 10 3 ) than between Galápagos-Peru comparisons (average dxy = 3.6 x 10 3 ; Figure 2C), and samples showing low genome-wide divergence were clustered in central Ecuador (similar patterns were observed using FST; SI Appendix, Table S10). To investigate potential source localities for invasive populations at a finer scale, we implemented the software Locator (Battey et al, 2020) which uses a machine learning algorithm to predict sample origins from genotype data. Locator predictions indicated 2 to 3 source regions for Galápagos PIM, although the exact locations varied across runs and depended on which island PIM collections were considered (SI Appendix, Fig.…”