Machine Learning for Prediction of Antibiotic Resistance

A team of researchers from University of California San Francisco and the Centers for Disease Control has developed a machine learning system that can predict antibiotic resistance patterns in bacterial infections with over 90% accuracy within hours, rather than the days traditionally required for culture-based testing. The system, called ARGnet (Antibiotic Resistance Genomic Network), was trained on whole-genome sequencing data from more than 10,000 clinical bacterial isolates collected from hospitals across five continents. By analyzing patterns in bacterial genomes, the algorithm can identify both known resistance genes and previously unrecognized genetic signatures associated with resistance. In validation testing at three major medical centers, ARGnet accurately predicted resistance to 39 common antibiotics across 12 bacterial species. The system was particularly effective with hard-to-treat pathogens like Pseudomonas aeruginosa and Acinetobacter baumannii, where it achieved accuracy rates of 92% and 94% respectively. Perhaps most impressively, the algorithm can generate predictions from nanopore sequencing data in as little as 4 hours from sample collection, potentially enabling same-day targeted antibiotic therapy. Several hospitals are now implementing the system in pilot programs, which could help preserve the effectiveness of existing antibiotics by ensuring they're only used when likely to be effective.