Reliable estimation of tree branch lengths using deep neural networks

Area of Science:

• Computational Biology
• Evolutionary Biology
• Machine Learning

Background:

• Phylogenetic trees visualize evolutionary history, incorporating both branching patterns and evolutionary distances (branch lengths).
• Estimating accurate branch lengths is crucial for understanding evolutionary relationships but can be challenging in certain parameter spaces.
• Machine learning (ML) shows promise for improving accuracy and efficiency in phylogenetic inference.

Purpose of the Study:

• To explore the potential of ML models, specifically deep learning frameworks, for predicting branch lengths in phylogenetic trees.
• To evaluate the performance of these ML models in estimating branch lengths on fixed tree topologies.

Main Methods:

• Developed and tested several deep learning frameworks to estimate branch lengths.
• Utilized multiple sequence alignments or their representations as input for the models.
• Compared deep learning performance against established methods like maximum likelihood and Bayesian inference.

Main Results:

• Deep learning methods demonstrated superior performance in difficult branch length parameter spaces.
• ML models were more efficient and accurate than maximum likelihood inference.
• Neural networks achieved accuracy comparable to Bayesian approaches, excelling at inferring long branches for distantly related taxa.

Conclusions:

• Deep learning offers a powerful tool for accurate and efficient phylogenetic branch length estimation.
• These findings represent a significant advancement towards reliable phylogenetic inference using ML.
• The study highlights ML's potential to overcome limitations of traditional phylogenetic methods.