disperseNN2: a neural network for estimating dispersal distance from georeferenced polymorphism data

Area of Science:

• Population Genetics
• Bioinformatics
• Machine Learning

Background:

• Organismal dispersal significantly influences spatial genetic variation.
• Accurate estimation of dispersal distance is crucial for understanding population structure and evolution.
• Existing methods may lack efficiency or applicability for non-model organisms.

Purpose of the Study:

• To introduce disperseNN2, a novel deep learning tool for estimating mean per-generation dispersal distance.
• To leverage georeferenced genetic polymorphism data for dispersal estimation.
• To provide a user-friendly tool for diverse biological systems.

Main Methods:

• Developed a neural network (disperseNN2) for feature extraction from genotype data.
• Integrated geographic information with genetic data within the neural network architecture.
• Utilized unphased single nucleotide polymorphisms (SNPs) as input, suitable for non-model organisms.

Main Results:

• disperseNN2 demonstrated superior performance compared to a state-of-the-art deep learning method lacking explicit spatial information.
• Achieved a 33% reduction in mean relative absolute error with 10 individuals and a 48% reduction with 100 individuals.
• The tool is particularly effective for organisms with sparse genomic resources.

Conclusions:

• disperseNN2 offers a powerful and accessible method for estimating dispersal distances from genetic data.
• The tool enhances the study of spatial genetics, especially in systems with limited genomic resources.
• Open-source availability promotes broader application and further development in evolutionary biology research.