Unifying phylogenetic traversal and deep learning to guide tree exploration.

Area of Science:

• Computational Biology
• Bioinformatics
• Machine Learning

Background:

• Deep learning (DL) has revolutionized many scientific fields but has yet to significantly impact phylogenetic inference.
• Current phylogenetic inference methods can be computationally intensive, especially for large datasets.
• There is a need for more efficient and accurate methods for reconstructing evolutionary relationships.

Purpose of the Study:

• To develop a novel approach combining deep learning with established phylogenetic algorithms for more efficient inference.
• To predict whether each edge of a phylogenetic tree belongs to a maximum parsimony tree.
• To lay the groundwork for future applications of deep learning in phylogenetic inference.

Main Methods:

• A deep learning model was developed, utilizing a recurrent neural network (RNN).
• The RNN was trained on the output of a phylogenetic dynamic programming algorithm applied to sequence alignments, not raw alignments.
• The model learns features from phylogenetically processed data to inform local tree search and classify tree edges.

Main Results:

• The proposed model achieved high-quality predictions for edge classification in maximum parsimony trees.
• The approach demonstrated effectiveness on both simulated and empirical datasets across various tree sizes.
• The method successfully addresses the NP-complete problem of phylogenetic tree reconstruction.

Conclusions:

• This novel deep learning approach enhances phylogenetic inference efficiency by leveraging phylogenetically informed data.
• The model's ability to predict tree edges represents a significant step towards practical DL applications in phylogenetics.
• This work paves the way for more scalable and accurate reconstruction of evolutionary histories.