AI helps discover new antibiotic candidates for drug-resistant gonorrhea

New research has demonstrated how AI can help identify novel antibiotic candidates for gonorrhea, one of the world’s most persistent and rapidly evolving infectious diseases.

Published in scientific translational medicinethis study shows that deep learning models can successfully navigate through millions of chemical structures to accurately identify compounds with activity against. gonorrheaincluding strains resistant to multiple existing antibiotics. The findings point to a potential new strategy to replenish the dwindling antibiotic pipeline as resistant bacteria continue to erode standard treatments.

Gonorrhea is the second most commonly reported sexually transmitted disease in the world, with tens of millions of cases reported each year. More than 600,000 infections are reported annually in the United States alone. Although often treatable, if left untreated, this infection can lead to serious complications, including infertility in both men and women, pelvic inflammatory disease, and increased susceptibility to HIV. In rare cases, if the infection reaches the bloodstream, it can cause life-threatening conditions such as meningitis, sepsis, and heart complications.

It is the speed of infection, rather than the infection itself, that is a growing clinical challenge. gonorrhea Resistance to antibiotics develops. Even newly introduced treatments, such as zoliflodacin and gepotidacin, are already expected to face resistance over time.

“We have seen this cycle of resistance development time and time again, within just five to 10 years after the deployment of first-line drugs,” Melis Anatar, lead author of the study and associate director of the Clinical Microbiology Laboratory at Massachusetts General Hospital, said in a press release. “To win this ongoing arms race, we need new antibiotics to fill our pipeline.”

The new research, led by James Collins from Harvard’s Wyss Institute, the Massachusetts Institute of Technology, and the Broad Institute, along with Anatal, Jackie Valeri, Majid Modareshi, and others, investigates how machine learning could accelerate pipelines.

Two AI studies conducted in parallel

This study was conducted in parallel with a second closely related effort led by Aarti Krishnan, Anahtar, Veleri, and Collins, published in 2025. cell. on the other hand, scientific translational medicine This research focuses on using deep learning to search vast existing chemical libraries for promising antimicrobial compounds. cell In this paper, we investigated whether generative AI could go a step further and design entirely new antibiotic molecules from scratch or from minimal chemical fragments.

The gonorrhea model described in scientific translational medicine The paper is the same one used to record chemical fragments. de novo-Compounds produced gonorrhea in cell “The two papers are closely related,” Anatal said. DDN. ” cell The essence of the paper is to apply the model to something. de novo design effort. Together, they demonstrate how a single deep learning framework can be used in a variety of ways to identify new antibiotic candidates. ”

Mapping chemical space with machine learning

For many years, gonorrhea has been treated with antibiotics that act on a variety of bacterial targets. Penicillins weaken cell walls, tetracyclines inhibit protein synthesis, quinolones target DNA replication, and azithromycin blocks ribosome function. Even the two newest drugs in development act on the same family of enzymes, type 2 topoisomerases, that fluoroquinolones target, although their binding sites are different.

The problem, Anatal explained, is that drugs designed to fit neatly into specific binding pockets are often highly vulnerable to resistance. “Just one or two point mutations are enough to completely prevent binding,” she said.

To circumvent this pattern and find new targets, the team trained a model based on phenotypic data rather than using traditional target-based approaches. The research team tested the ability of 38,650 small molecules to inhibit proliferation. N. gonorrhoeae in vitroWe generate large experimental data sets that capture whether compounds can actually kill or suppress pathogens inside cells. These results were used to train a neural network that can predict antibacterial activity from chemical structure alone.

We’re not looking for the next bleach. We are looking for candidates that target something specific to bacterial cells.

—Melis Anatal, Massachusetts General Hospital

However, phenotypic screening comes with its own risks. “You could end up selecting compounds that are generally only toxic,” Anatal said. To address this, both studies used a human cytotoxicity counterscreen to apply rigorous filtering, first computationally and then experimentally, to remove compounds that nonspecifically kill cells. “We’re not looking for the next bleach,” she said. “We want candidates that target something specific to bacterial cells.”

This strategy worked for both papers. in cell As a result of their research, the research team identified two novel compounds called NG1 and DN1 as particularly promising. NG1 showed narrow-spectrum activity against pathogens gonorrheawhereas DN1 is designed to target targets. Staphylococcus aureus However, it showed broader activity against gonorrhea.

Importantly, both compounds acted through mechanisms different from commonly used antibiotics and were able to reduce bacterial burden in animal models of infection. Beyond immediate antibacterial effects, the researchers highlighted that both NG1 and DN1 are within chemical families suitable for further medicinal chemistry optimization.

Similarly, scientific translational medicine Two lead compounds were discovered in the study. One, MP20, does not appear to have a single clear protein target, but instead appears to increase bacterial membrane permeability through an as-yet-unknown mechanism. The second, known as A1, has been shown to bind to alanine racemase, an enzyme essential for bacterial cell wall synthesis but not previously targeted by small-molecule antibiotics for gonorrhea.

This indicates that both approaches are viable for discovering antibiotic candidates that may be missed by traditional discovery strategies.

From calculations to biological systems

Anatal said he is particularly excited about testing the compound in an organ-on-a-chip model. “Especially in mouse models of vaginal infections, we often just apply a compound and observe the end result,” she says. “It’s very difficult to investigate what’s actually going on at the organizational level.”

Although still in its early stages, a vaginal organ model on a chip developed by Donald Ingber’s lab at the Wyss Institute has proven promising. Researchers were initially content to just observe that gonorrhea It attaches to, invades, and acts appropriately within the human cellular environment. This platform consists of a porous polymer membrane seeded with multiple layers of human vaginal epithelial cells on one side and a layer of human fibroblasts on the other side. Compounds can be introduced from above to mimic local administration or from below to model systemic delivery.

Using this system, the research team showed that MP20 can cross epithelial barriers, cross tight junctions, and die. gonorrhea Located in the vaginal lumen. “That was really interesting to see,” Anatal said, noting that a compound’s permeability is often one of the biggest unknowns in antibiotic development. “Sometimes when something goes wrong, you don’t know why. In this case, it worked. I was very excited.”

Beyond proof of effectiveness, Anahtar considers the Organ-on-a-Chip model to be a powerful tool for troubleshooting and optimization. Although this technology is still specialized and only available in a few laboratories, commercial versions of vaginal chips are emerging. Because these systems rely on human cells rather than animal physiology, they may help avoid some of the translational pitfalls that arise when compounds have to be optimized for mice and then redesigned for humans.

“Although it is still early days, we hope that these models will allow deeper mechanistic elucidation and help us understand why certain compounds work well, why others do not, and how to choose between multiple analogs for very specific clinical indications,” she said.

Toward a new antibiotic pipeline

This research builds on a growing body of evidence that machine learning approaches can provide real momentum in antibiotic discovery. In previous research, Collins and colleagues used a similar AI-driven strategy to identify several promising drug candidates, including halicin and abausin. Haricin was originally studied as a potential diabetes drug, but it turned out to be highly effective against diabetes. clostridioides difficile Pan resistance Acinetobacter baumannii Infection in mouse models. In contrast, avausin showed narrow-spectrum activity against: A. Baumani — a major advantage for more selective antibiotic treatment.

By combining deep learning with large-scale chemical screens and physiologically relevant models, this approach provides a way to systematically explore antibiotic candidates at a scale and speed that was previously unfeasible.

As antibiotic resistance continues to increase globally, AI-powered discovery strategies are likely to become increasingly central to maintaining the effectiveness of future treatments.