Selecting informative subsets of sparse supermatrices increases the chance to find correct trees

Area of Science:

• Phylogenomics
• Computational Biology
• Systematics

Background:

• Phylogenomic analyses often use matrices with extensive missing data, potentially compromising tree inference accuracy.
• Current methods for selecting taxa and genes rely heavily on data coverage, potentially overlooking actual phylogenetic signal.
• This can lead to suboptimal data matrices that do not maximize phylogenetic signal.

Purpose of the Study:

• To develop and evaluate a heuristic approach for selecting informative genes in phylogenomic datasets.
• To improve the accuracy and robustness of phylogenetic tree inference from sparse matrices.
• To provide a method that considers both data coverage and phylogenetic signal content.

Main Methods:

• Developed a software tool named 'mare' implementing a heuristic approach.
• Assessed gene information content using a measure combining potential phylogenetic signal and data coverage.
• Reduced supermatrices to informative submatrices using a hill-climbing procedure.

Main Results:

• Maximum Likelihood tree reconstructions failed with simulated sparse matrices (50 taxa × 50 genes, 10-30% data coverage).
• The proposed method increased the recovery of correct partial trees more than tenfold compared to standard methods.
• Application to an empirical vertebrate dataset showed that selecting data subsets based on information content improved tree support.

Conclusions:

• Sparse supermatrices can be formally reduced to improve phylogenetic inference.
• The developed heuristic approach outperforms traditional methods relying solely on data coverage.
• This method provides a robust basis for selecting informative data subsets in phylogenomics.