A LASSO-based approach to sample sites for phylogenetic tree search

Area of Science:

• Computational Biology
• Bioinformatics
• Phylogenetics

Background:

• Modern phylogenetic analyses increasingly use large-scale, full-genome sequences (>100,000 sites).
• Phylogenetic reconstruction with these large datasets is computationally intensive, often requiring powerful computer clusters.
• Existing alignment trimming tools offer limited size reduction and may negatively impact tree accuracy.

Purpose of the Study:

• To develop an artificial intelligence (AI)-based approach for efficient phylogenetic analysis of large sequence alignments.
• To identify an optimal subset of sites for accurate log-likelihood computation.
• To reduce computational time for phylogenetic tree searches while maintaining accuracy.

Main Methods:

• An AI-based approach using regularized Lasso regression to select informative sites.
• Training a model to optimize log-likelihood prediction accuracy with a constraint on the number of sites.
• Computing the tree likelihood based on a selected subset of sites.

Main Results:

• Accurate approximation of the full dataset's tree likelihood using only 5% of the sites.
• Substantial decrease in running time during tree search when using the Lasso-based approximation.
• Retained tree-search performance comparable to methods using the entire dataset.

Conclusions:

• The proposed AI method effectively reduces computational demands in phylogenetic analysis.
• This approach offers a significant speed-up for large-scale phylogenetic reconstructions.
• The method provides an accurate and efficient alternative for analyzing extensive sequence data.