Identification of Molecular Compounds Targeting Bacterial Propionate Metabolism with Topological Machine Learning

Abstract

This study demonstrates the effective integration of comparative protein sequence analysis with novel topological machine learning methods to tackle a key issue in computational biology: identifying potential inhibitor compounds for methylcitrate dehydratase--an enzyme essential to the methylcitrate pathway in bacteria and fungi. While many ML models have proven effective on benchmark datasets, we applied these techniques specifically to discover compounds for this target protein. This pathway, crucial for metabolizing propionic acid, is essential in pathogenic bacteria for utilizing host-derived lipids and amino acids, making it a promising antimicrobial target. Inefficient removal of propionate can lead to toxic accumulation, threatening microbial survival. In our research, we utilize the latest methods in topological machine learning, multiparameter persistence, to transform the molecular structure of potential compounds into topological vectors for ligand-based drug discovery. By applying our tailored topological model, we prioritized 15 compounds with promising characteristics for inhibiting the active site of methylcitrate dehydratase. Computational molecular docking simulations reveal that these identified compounds interact with key amino acid residues critical to the enzymes function. Our findings underscore the power of integrating topology-based modeling with comparative sequencestructure-function analysis and ligand docking. Considering compounds geometric fit, energy, and interaction profiles enhances predictions and guides the design of optimized new compounds. This refined approach offers a solid foundation for lead optimization, representing the most promising molecular scaffolds for modification, advancing compound discovery, and providing valuable insights into binding interactions for further experimental validation and potential drug development. Our code is available at https://github.com/AstritTola/Molecular-Compounds-Targeting