Evolutionary Sparse Learning for Phylogenomics.

Area of Science:

• Genomics
• Computational Biology
• Machine Learning

Background:

• Phylogenomics traditionally uses complex models of sequence evolution.
• Identifying influential genomic loci for phylogenetic hypotheses is challenging.
• Existing methods may not efficiently integrate diverse data types or annotations.

Purpose of the Study:

• Introduce Evolutionary Sparse Learning (ESL), a novel supervised machine learning approach for phylogenomics.
• Develop a method that directly uses genomic loci as parameters and incorporates sparsity constraints.
• Enable the combination of diverse data types and biological annotations within phylogenetic modeling.

Main Methods:

• ESL employs the Least Absolute Shrinkage and Selection Operator (LASSO) for feature selection of genomic loci.
• Models are built based on the concordance of sequence variation with evolutionary hypotheses, bypassing traditional parameters.
• Proposes novel sparsity scores: positional, gene, function, and hypothesis.

Main Results:

• ESL effectively selects the most important genomic loci relevant to phylogenetic hypotheses or trait evolution.
• Demonstrates the ability to integrate various molecular and non-molecular data, along with locus annotations.
• Achieves fast computation times and a small memory footprint.

Conclusions:

• ESL offers a complementary computational framework to traditional phylogenomic methods.
• It excels at identifying influential loci and sequences for hypothesis testing.
• ESL has the potential to democratize big data analytics in evolutionary genomics and enhance scientific rigor.