Estimating Genome-Wide Phylogenies Using Probabilistic Topic Modeling.

Area of Science:

• Computational Biology
• Evolutionary Genetics
• Bioinformatics

Background:

• Genome-wide data offers unprecedented resolution for inferring evolutionary history.
• Existing phylogenetic methods face computational limitations with large genomic datasets.
• Alignment-free approaches are emerging as a way to overcome these challenges.

Purpose of the Study:

• To develop and implement an efficient, alignment-free computational method for inferring genome-wide phylogenies.
• To introduce the Python package TopicContml for phylogenetic analysis.
• To assess the performance and robustness of the new method on various datasets.

Main Methods:

• Developed an alignment-free phylogenetic inference method utilizing probabilistic topic modeling (Latent Dirichlet Allocation or LDA).
• Extracted 'topic' frequencies from k-mers derived from multilocus DNA sequences.
• Integrated TopicContml with the Contml program from the PHYLIP package for species tree generation.
• Employed a bootstrap procedure to assess the uncertainty of estimated clade relationships.

Main Results:

• TopicContml demonstrated efficiency and statistical robustness on simulated datasets, including those with gaps.
• The method was successfully applied to three distinct biological datasets: Australian bird populations, mammal species, and PacBio sequences from bird species.
• Empirical results and simulations indicate reliable phylogenetic inference capabilities.

Conclusions:

• TopicContml provides an efficient and statistically robust alignment-free approach for genome-wide phylogenetic inference.
• The method effectively handles diverse and complex biological sequence data, overcoming computational constraints.
• This tool advances the field of phylogenomics by enabling rapid evolutionary history reconstruction.