Leveraging graphical model techniques to study evolution on phylogenetic networks.

Area of Science:

• Evolutionary biology
• Computational biology
• Phylogenetics

Background:

• Trait evolution is often modeled using Markov processes on phylogenetic trees.
• Phylogenetic networks, which include reticulations like hybridization, pose computational challenges for likelihood calculations.
• Existing algorithms for trees are not applicable to networks.

Purpose of the Study:

• To develop efficient computational methods for trait evolution models on phylogenetic networks.
• To adapt graphical model techniques for phylogenetic network analysis.
• To reduce the computational cost of likelihood and parameter inference for complex evolutionary histories.

Main Methods:

• Reformulating trait evolution models on phylogenetic networks as graphical models.
• Applying belief propagation algorithms to these graphical models.
• Focusing on linear Gaussian models for continuous traits.
• Developing exact and approximate likelihood and gradient calculations.

Main Results:

• Demonstrated that phylogenetic network models can be represented as graphical models.
• Showcased the application of belief propagation for efficient likelihood and gradient computations.
• Proved novel results for efficient parameter inference in certain models.
• Highlighted the potential of approximate likelihood methods to significantly reduce computational costs.

Conclusions:

• Graphical models and belief propagation offer a powerful framework for analyzing trait evolution on phylogenetic networks.
• These methods significantly improve computational efficiency, especially for complex networks with reticulations.
• This approach bridges graphical models and phylogenetic methods, opening new avenues for evolutionary research.