Phylogenetic inference using generative adversarial networks.

Area of Science:

• Computational Biology
• Evolutionary Biology
• Machine Learning

Background:

• Phylogenetic inference, crucial for understanding evolutionary history, faces computational challenges due to vast model spaces.
• Supervised machine learning methods require extensive data across these spaces, limiting their application to simpler phylogenetic problems (e.g., unrooted quartets).

Purpose of the Study:

• To explore the potential of generative adversarial networks (GANs) to address the computational limitations in phylogenetic inference.
• To develop and evaluate a GAN-based tool, phyloGAN, for inferring phylogenetic relationships.

Main Methods:

• Developed phyloGAN, a GAN framework utilizing an evolutionary model as the generator to create data for phylogenetic inference.
• Tested phyloGAN on concatenated alignments and gene alignments for up to 15 taxa, assessing its ability to infer phylogenetic trees.
• Explored inference both with and without considering gene tree heterogeneity.

Main Results:

• phyloGAN demonstrated relatively low error rates in inferring phylogenetic relationships for tested taxa.
• The study identified slow run times and training instabilities as areas for improvement in the current phyloGAN architecture.
• Performance was evaluated for up to 15 taxa in concatenation and 6 taxa when accounting for gene tree heterogeneity.

Conclusions:

• Generative adversarial networks present a viable machine learning approach to navigate complex model spaces in phylogenetics.
• phyloGAN offers a new computational tool for evolutionary inference, though further architectural development is needed for enhanced efficiency and stability.
• The study highlights the potential of GANs to advance machine learning applications in phylogenetics.