MyESL: Sparse learning in molecular evolution and phylogenetic analysis.

Area of Science:

• Genomics
• Computational Biology
• Machine Learning

Background:

• Evolutionary sparse learning (ESL) applies supervised machine learning, specifically LASSO, to model relationships between hypotheses and genomic feature variation.
• Existing sparse group LASSO software struggles with the scale of genome-wide sequence data, necessitating specialized tools.

Purpose of the Study:

• To introduce MyESL, an open-source software package designed for efficient and flexible ESL analyses on large genomic datasets.
• To provide tools for pre-processing, flexible LASSO model building, and post-processing of results for evolutionary genomics.

Main Methods:

• MyESL utilizes a C++ core for computationally intensive model building (with or without group sparsity) and Python for pre/post-processing.
• Input data, including binary responses or phylogenetic trees and sequence alignments, are transformed into a suitable format for LASSO analysis.
• The software handles continuous and binary features, converting sequence alignments into binary one-hot encoded matrices.

Main Results:

• MyESL successfully builds evolutionary models rapidly and efficiently on standard desktop computers.
• Empirical genome-scale dataset analyses demonstrate MyESL's capability where other packages failed due to resource limitations.
• The software generates user-friendly text and graphical outputs for downstream interpretation.

Conclusions:

• MyESL addresses the limitations of existing software for ESL on large-scale genomic data.
• It offers a computationally efficient and accessible solution for evolutionary and functional genomics research.
• MyESL is available for Linux, Windows, and macOS, facilitating integration into diverse bioinformatic pipelines.