Efficient identification of phylogenetically informative alignment sites via sparse learning.

Area of Science:

• Phylogenetics and Evolutionary Biology
• Computational Biology
• Genomics

Background:

• Accurate phylogenetic tree reconstruction relies on identifying phylogenetically informative sites in multiple sequence alignments.
• Current methods often depend on predefined topologies or heuristics, limiting their applicability and interpretability.

Purpose of the Study:

• To develop a topology-agnostic framework for quantifying site-wise phylogenetic information.
• To identify the minimal subset of sites crucial for phylogenetic signal using sparse learning.

Main Methods:

• Employed sparse learning via Lasso (Least Absolute Shrinkage and Selection Operator) regression.
• Modeled site log-likelihoods as predictors of tree likelihood across random topologies.
• Validated using simulated and empirical mammalian datasets.

Main Results:

• Lasso-selected sites produced tree topologies nearly identical to those from full alignments.
• An entropy-based proxy effectively approximated Lasso results for computational efficiency.
• Demonstrated the identification of a minimal subset of phylogenetically informative sites.

Conclusions:

• Sparse learning offers a principled, scalable, and practical method for assessing and optimizing phylogenetic data.
• The developed framework provides an objective metric for phylogenetically informative sites.
• This approach enhances efficiency and accuracy in phylogenomic analyses.