Machine learning discovers new antibiotics to fight resistant infections
The continued proliferation of antibiotic-resistant gonorrhea poses a growing threat to global public health, necessitating innovative solutions in antimicrobial drug discovery. Gonorrhea, caused by Neisseria gonorrhoeae, is one of the most common sexually transmitted diseases worldwide, with more than 600,000 cases reported annually in the United States alone. If left untreated, it can lead to severe reproductive health complications such as infertility and pelvic inflammatory disease, as well as increasing the risk of HIV infection. A particularly difficult challenge is the rapid evolution of pathogen resistance to newly introduced antibiotics, rendering traditional treatment strategies increasingly ineffective.
Recently, new oral antibiotics such as zoliflodacin and gepotidacin have been introduced, making them the first new class of antibiotics used to treat uncomplicated urogenital gonorrhea in more than 30 years. However, these drugs cannot ignore the adaptability of Neisseria gonorrhoeae. Historical trends indicate that resistance often surfaces within 5 to 10 years of widespread use. This evolutionary arms race highlights the urgent need for continued antibiotic innovation to replenish drug development pipelines and maintain clinical efficacy against this resilient pathogen.
A pioneering study published in Science Translational Medicine introduces a machine learning-driven approach to gonorrhea-specific antibiotic discovery. A research team led by Dr. James Collins of the Wyss Institute for Bioinspired Engineering, Harvard University, Massachusetts Institute of Technology, and the Broad Institute utilized deep learning algorithms to interrogate a vast chemical library of compounds exhibiting novel antimicrobial activity. This hypothesis was based on the premise that unconventional chemical structures that may target rare or previously unstudied bacterial pathways may reduce the likelihood of resistance development.
To establish a functional predictive model, the researchers first screened a comprehensive set of approximately 38,650 small molecules for their inhibitory effects on Neisseria gonorrhoeae growth in vitro. This assay data trained a deep learning platform that was able to identify chemical features that predicted anti-gonococcal activity beyond structural similarity to existing antibiotics. Validation experiments confirmed the model’s ability to identify drug-like molecules with antimicrobial potential, including compounds that are structurally distinct from traditional antibiotic classes.
Subsequent in silico screening was expanded to a vast virtual chemical library containing approximately 6 million candidates. From this virtual screen, 213 promising compounds emerged, which were subjected to rigorous in vitro growth inhibition assays and toxicity evaluation. This filtering process ultimately revealed two compounds that exhibited significant selectivity and potent inhibitory potency against multidrug-resistant N. gonorrhoeae strains. Remarkably, these compounds also show the emergence of resistance at low frequencies and exhibit durable antimicrobial effects.
Delving deeper into the mechanism of action, proteomic analysis revealed that the most promising compound, called A1, is an aminothiazole derivative with a novel target: alanine racemase. This enzyme catalyzes the conversion of L-alanine to D-alanine, an essential precursor in bacterial peptidoglycan cell wall biosynthesis. Inhibiting alanine racemase disrupts cell wall architecture and compromises bacterial integrity. Although inhibition of cell wall biosynthesis is a known antibiotic strategy, direct targeting of alanine racemase with small molecules is unprecedented and represents an innovative therapeutic avenue for gonorrhea.
These encouraging molecular insights led research to physiological evaluation of antimicrobial effects within human relevant tissues. The researchers used a microfluidic organ-chip model of the human vagina developed by co-author Donald Ingber’s team to simulate a natural infection environment. They demonstrated that one of the key compounds, MP20, significantly reduced gonococcal colonization of vaginal epithelial cells within this engineered system. Complementing this, a murine vaginal infection model validated the in vivo potential of the alanine racemase inhibitor A1, where intravaginal administration significantly reduced bacterial load over multiple treatments within 24 hours.
Despite these promising preclinical findings, the authors highlight the need for further medicinal chemistry optimization and detailed mechanistic studies to improve the compound’s efficacy, pharmacokinetics, and safety profile before clinical application. However, deep learning-based discovery platforms signal a powerful paradigm shift in integrating artificial intelligence with high-quality biological datasets and human-relevant models to accelerate antibiotic innovation.
This work also exemplifies a broader trend at the interface of computational biology, chemical science, and tissue engineering, with AI-driven approaches unlocking a vast chemical space previously inaccessible with traditional methodologies. The ability to rapidly identify and characterize entirely new bioactive compounds increases the likelihood of gaining an advantage in the lasting fight against antimicrobial resistance.
This multidisciplinary research, supported by a collaborative network that includes the Defense Threat Reduction Agency, the National Institutes of Health, the Swiss and Swedish Research Foundation, and philanthropic organizations such as the Bill and Melinda Gates Foundation, highlights the critical role of sustained funding and cross-sector partnerships in addressing urgent global health crises.
Finally, the fusion of machine learning and advanced human tissue models offers a glimmer of hope in the fight against drug-resistant pathogens like Neisseria gonorrhoeae. As resistance dynamics continue to outpace traditional drug development, such integrated and innovative approaches are poised to redefine antibiotic discovery and usher in new frontiers in infectious disease treatment.
Research theme: animal
Article title: Discovery of antibiotics effective against gonorrhea using deep learning
News publication date: June 17, 2026
Web references:
Harvard University Wyss Institute
Scientific Translational Medicine Journal
References:
Valeri, J., Modaresi, M., Anahtar, M., Collins, J. et al. Discovery of antibiotics effective against gonorrhea using deep learning. Science Translational Medicine (2026).
Image credits: Wyss Institute for Bioinspired Engineering, Harvard University
keyword
machine learning, artificial intelligence, sexually transmitted diseases, infectious diseases, antibiotic resistance, antibiotic activity, computational biology, vagina, mouse models, tissue engineering, compounds, bioactive compounds, chemical modeling, computational chemistry, antibiotics
Tags: Drug Discovery with AI Antibiotic Resistance Treatment of Gonorrhea Antibiotic Development Pipeline Fighting Multidrug-Resistant Infections Evolutionary Resistance of Bacteria Global Public Health Antibiotic Challenge Gonorrhea Reproductive Health Complications Antibiotic Discovery with Machine Learning Machine Learning in Infectious Diseases Novel Antibiotics for Gonococcal Resistance STIszoliflodacin and gepotidacin
