LB 26-221 - Estimating maps of population density and dispersal from genetic data using machine learning

Population density and dispersal of organisms across landscapes shape patterns of genetic variation within species, such that it is possible to infer demographic properties from nucleotide sequencing data. Here we present a new inference method that uses geographically distributed genotype data in combination with deep neural networks to estimate spatial maps of dispersal and population density. The maps estimated by our program convey both the magnitude of dispersal and density, and how these parameters vary over the landscape. Our neural network trains on simulated input and output pairings, where the input consists of genotypes and sampling locations generated from a continuous space population genetic simulator, and the output is the ground-truth map of demographic parameters that went into each simulation. We benchmark our approach against existing methods and discuss qualitative differences between the different approaches. In particular, our program is unique because it uses single nucleotide polymorphism data to infer the magnitudes of dispersal and density. This input data type is available for many non-model species and is becoming accessible for a growing number of unconventional study organisms. Additionally, our method can be used with small sample sizes, although spatial precision scales accordingly, further reducing the cost of obtaining spatial demographic estimates via sequencing. We apply our program to empirical data from North American gray wolves and discuss the associated workflow and findings. Genetic tools like ours complement other, direct methods for improving our understanding of population demography, and we believe will provide a valuable tool for conservation genetics applications.