Tulane University scientists have developed a machine learning model that more accurately detects antibiotic resistance in deadly bacteria—marking a major step toward faster, more effective treatments.
Image Credit: nobeastsofierce/Shutterstock.com
Drug-resistant infections—particularly those caused by bacteria like tuberculosis and staph—are an escalating global health concern. These infections are tougher to treat, often require more costly or toxic medications, and are linked to longer hospital stays and higher mortality rates. In 2021 alone, 450,000 people developed multidrug-resistant tuberculosis, and treatment success rates dropped to just 57 %, according to the World Health Organization.
The new approach, developed by researchers at Tulane, identifies genetic markers of resistance in Mycobacterium tuberculosis and Staphylococcus aureus using artificial intelligence.
The study, published in Nature Communications, introduces the Group Association Model (GAM), a machine learning tool designed to pinpoint genetic mutations associated with drug resistance. Unlike traditional methods — which often misattribute unrelated mutations as resistance markers — GAM doesn’t depend on prior knowledge of resistance mechanisms. This flexibility allows it to uncover previously unidentified genetic changes.
Think of it as using the bacteria’s entire genetic fingerprint to uncover what makes it immune to certain antibiotics. We’re essentially teaching a computer to recognize resistance patterns without needing us to point them out first.
Tony Hu, Ph.D, Study Senior Author and Weatherhead Presidential Chair and Director, Biotechnology Innovation, Center for Cellular & Molecular Diagnostics, Tulane University
Existing diagnostic methods, like those used by the WHO, either take too long—such as culture-based testing—or miss rare mutations, as seen in some DNA-based tests. Tulane’s model addresses both limitations by analyzing whole genome sequences and comparing bacterial strains with different resistance profiles to identify mutations consistently linked to resistance.
The research team applied GAM to more than 7000 M. tuberculosis strains and nearly 4000 S. aureus strains, uncovering key mutations tied to drug resistance. The model not only matched or exceeded the accuracy of the WHO’s resistance database but also significantly reduced false positives—a crucial improvement since misclassifying bacteria as resistant can lead to ineffective or unnecessary treatments.
Current genetic tests might wrongly classify bacteria as resistant, affecting patient care. Our method provides a clearer picture of which mutations actually cause resistance, reducing misdiagnoses and unnecessary changes to treatment.
Julian Saliba, Study Lead Author and Graduate Student, Center for Cellular & Molecular Diagnostics, Tulane University
The model also demonstrated strong performance with limited or incomplete data. In validation studies using clinical samples from China, the AI-powered system outperformed WHO-based methods in predicting resistance to several first-line antibiotics.
This is a critical advancement; detecting resistance early helps physicians tailor treatment before an infection worsens or spreads.
Because GAM doesn’t rely on expert-defined resistance rules, it has potential applications beyond TB and staph. It could be used to track antibiotic resistance in other bacterial infections—or even in agriculture, where resistance threatens crops and food systems.
“It is vital that we stay ahead of ever-evolving drug-resistant infections. This tool can help us do that,” added Saliba.
Journal Reference:
Saliba, G, J., et al. (2025) Enhanced diagnosis of multi-drug-resistant microbes using group association modeling and machine learning. Nature Communications. doi.org/10.1038/s41467-025-58214-6