Phylogenetic Methods Meet Deep Learning.

Area of Science:

• Computational Biology
• Genomics
• Evolutionary Biology

Background:

• Deep learning (DL) applications in phylogenetics are emerging but face challenges with complex data.
• Current DL studies often focus on small, four-taxon trees, serving mainly as proofs of principle.
• Traditional phylogeny reconstruction methods are well-established but can be computationally intensive.

Purpose of the Study:

• To provide a perspective on the application of deep learning in phylogenetics.
• To introduce prevalent deep learning architectures relevant to phylogenetic analysis.
• To highlight potential challenges and promising future research directions in DL for phylogenetics.

Main Methods:

• Utilizing advanced data encoding techniques like compact bijective ladderized vectors and transformers to handle large datasets.
• Exploring various deep learning architectures suitable for phylogenetic inference.
• Discussing the use and risks of simulation-based training data.

Main Results:

• New data encoding methods allow deep learning to analyze much larger phylogenetic trees and genomic datasets.
• Deep learning models show potential to perform comparably to traditional methods, with possibilities for significant computational cost reduction.
• Identified risks associated with simulation-based training data, emphasizing the need for reproducibility and robustness.

Conclusions:

• Deep learning offers a powerful complement to traditional phylogeny reconstruction methods.
• DL can significantly aid phylogenetic analysis, particularly in computationally demanding tasks like model selection and branch support estimation.
• Future research should explore DL integration with population genetics and analysis of neighbor dependencies for enhanced phylogenetic insights.