Estimation of spatial demographic maps from polymorphism data using a neural network.

Area of Science:

• Population genetics
• Landscape genetics
• Computational biology

Background:

• Understanding how genetic variation is distributed across landscapes is crucial in population genetics.
• Heterogeneous population densities and dispersal barriers are known to influence genetic variation.
• Existing tools often struggle to account for these complex spatial factors.

Purpose of the Study:

• To develop a new inference method for estimating spatially heterogeneous population density and dispersal rates.
• To utilize geo-referenced single nucleotide polymorphism (SNP) data and deep neural networks for this purpose.
• To offer a tool that infers both the magnitude and spatial variation of demographic parameters.

Main Methods:

• A deep neural network was trained on simulated genotype and sampling location data.
• The network learned to predict demographic parameters (density, dispersal rate) from SNP data.
• The method was benchmarked against existing population genetics inference tools.

Main Results:

• The developed method successfully estimated spatially heterogeneous maps of population density and dispersal rate.
• Benchmarking revealed unique capabilities compared to existing methods, which often estimate relative migration or require specific genetic blocks.
• Application to North American grey wolf data yielded reasonable demographic parameter estimates, despite challenges from incomplete spatial sampling.

Conclusions:

• The new method provides a valuable tool for inferring landscape-level demographic parameters from SNP data.
• It complements existing methods in population genetics, ecology, and evolutionary biology.
• The open-source software facilitates applications in conservation and ecological studies.