Machine Learning Approaches Using High-Throughput Profiling Data for Antibiotic Discovery

Antibiotic-resistant bacterial infections continue to increase globally, creating an urgent need for new antibacterials with novel mechanisms of action. Early stages of antibiotic discovery are often limited by the difficulty of identifying compounds that act through previously unrecognized pathways and by challenges in determining their mechanisms. Machine learning (ML) integrated with high-throughput profiling now provides systematic approaches to overcome these barriers. Beyond initial hit discovery, multilayer profiling using morphological phenotyping, transposon sequencing, transcriptomics, proteomics, and metabolomics captures cellular responses that reflect the mechanisms of action. Because profiling data sets are typically high-dimensional and contain defined features and variables, ML can extract complex patterns associated with pathway-level responses and predict mechanisms for unknown compounds. In this review, we summarize current progress in high-throughput profiling and describe how ML applied to each data set can accelerate the identification of antibacterials with new mechanisms. These approaches accelerate the transition from large-scale compound screening to mechanistic validation and enable effective prioritization of lead compounds in early-stage antibiotic discovery.