Phylogeographic model selection using convolutional neural networks.

Area of Science:

• Evolutionary Biology
• Genomics
• Computational Biology

Background:

• Phylogeography utilizes genomic data for evolutionary insights.
• Advanced analytical tools are needed for complex demographic models.
• Deep learning, such as CNNs, remains underutilized in phylogeography.

Purpose of the Study:

• To assess the utility of CNNs for demographic model selection in phylogeography.
• To evaluate CNN performance in South American Norops lizards.
• To compare CNNs with Approximate Bayesian computation (ABC) for model selection.

Main Methods:

• Applied convolutional neural networks (CNNs) to analyze genomic data.
• Tested three demographic scenarios (constant, expansion, bottleneck) across four lineages.
• Evaluated 26 complex models incorporating gene flow and population size changes.
• Utilized Approximate Bayesian computation (ABC) for comparative analysis.

Main Results:

• CNNs achieved over 98% accuracy in identifying basic demographic scenarios.
• CNNs identified a single, best-fit demographic model with 87% accuracy from 26 complex models.
• Inferred demography suggests gene flow and population size changes shaped lizard genetic diversity.
• ABC analysis failed to identify a single best model from the complex set.

Conclusions:

• CNNs are a valuable and easily integrated tool for phylogeographic analysis.
• Deep learning enhances the ability to select complex demographic models.
• CNNs provide a more robust approach compared to ABC for complex phylogeographic questions.