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

Area of Science:

• Population genetics
• Computational biology
• Machine learning

Background:

• Dispersal is a key factor shaping spatial genetic variation.
• Accurately estimating dispersal distance is crucial for understanding population structure and evolution.
• Existing methods may struggle with non-model organisms or limited genomic data.

Approach:

• Introduced disperseNN2, a deep learning tool for estimating mean per-generation dispersal distance.
• Utilizes georeferenced polymorphism data, including unphased single nucleotide polymorphisms.
• Neural network performs feature extraction on genotype pairs combined with geographic information.

Key Points:

• disperseNN2 outperforms a state-of-the-art deep learning method by 33% (10 individuals) and 48% (100 individuals) in reducing mean relative absolute error.
• The tool is particularly effective for non-model organisms and systems with sparse genomic resources.
• Input data includes unphased, single nucleotide polymorphisms, making it broadly applicable.

Conclusions:

• disperseNN2 provides a powerful and accessible method for estimating dispersal in diverse organisms.
• The tool enhances our ability to study the genetic consequences of dispersal.
• Open-source availability facilitates wider adoption in ecological and evolutionary research.