Parameter Estimation from Phylogenetic Trees Using Neural Networks and Ensemble Learning.

Area of Science:

• Evolutionary Biology
• Computational Biology
• Phylogenetics

Background:

• Species diversification is driven by speciation and extinction.
• Estimating these rates from phylogenies is crucial but challenging.
• Current maximum likelihood estimation (MLE) methods have limitations with complex models and small phylogenies.

Purpose of the Study:

• To develop and evaluate a novel ensemble neural network approach for estimating diversification parameters from phylogenetic trees.
• To compare the performance of the neural network method against MLE and existing deep learning approaches.
• To assess the robustness of neural network methods in handling phylogenetic data.

Main Methods:

• An ensemble neural network combining dense neural networks, graph neural networks, and recurrent neural networks was developed.
• The network learns from graph representations, branching times, and summary statistics of phylogenies.
• Performance was evaluated against MLE and a convolutional network approach using simulated phylogenetic data.

Main Results:

• The ensemble neural network provides faster estimates than MLE and is less sensitive to tree size for constant-rate and diversity-dependent models.
• It performs comparably to existing convolutional network methods.
• Both the neural network approach and MLE struggle with parameter recovery under protracted birth-death processes.

Conclusions:

• The ensemble neural network is a promising, faster, and less biased alternative to MLE for estimating diversification parameters, especially when MLE is not feasible.
• Phylogeny size and the strength of evolutionary signals are key limitations for accurate parameter estimation.
• The method shows good performance when sufficient phylogenetic signal is present.