Machine Learning-Based Characterization of Bacillus anthracis Phenotypes from pXO1 Plasmid Proteins

Area of Science:

• Genomics and Bioinformatics
• Microbial Evolution
• Computational Biology

Background:

• The Bacillus anthracis pXO1 plasmid encodes key virulence factors and offers a model for studying bacterial evolution due to its slow evolutionary rate.
• Limited amino acid variation in Bacillus anthracis facilitates the detection of functionally relevant patterns within plasmid protein composition.

Purpose of the Study:

• To characterize pXO1 protein modules across diverse Bacillus anthracis lineages using advanced computational methods.
• To identify plasmid-encoded targets for assessing anthrax sublineage and understanding evolutionary relationships.

Main Methods:

• Application of embedding-based analyses and machine learning techniques.
• Generation of protein sequence embeddings and construction of phylogenies.
• Comparison of plasmid content with whole-genome variation and association rule mining.

Main Results:

• Plasmid protein composition revealed lineage-specific structures, diverging from whole-genome phylogenies.
• Association rule mining identified plasmid-encoded targets for sublineage assessment, highlighting functionally redundant modules.
• A conserved DNA replication module displayed both shared and lineage-specific features.

Conclusions:

• pXO1 plasmid protein modules contain evolutionarily informative signatures valuable for phylogeographic characterization of bacterial pathogens.
• The developed framework using protein language models can be extended to study other virulence plasmids and environmental pathogens.