In a groundbreaking advance against the escalating crisis of antibiotic resistance, researchers at MIT have harnessed artificial intelligence to design entirely new antibiotics capable of tackling two notorious drug-resistant bacteria.
The antibiotics can be used to treat Neisseria gonorrhoeae, the culprit behind gonorrhea, and methicillin-resistant Staphylococcus aureus (MRSA), a common cause of severe skin and bloodstream infections.
The study, published today in the journal Cell, comes at a critical time. Over the past 45 years, the FDA has approved only a handful of new antibiotics, most of which are mere tweaks on existing drugs.
Meanwhile, bacterial resistance has surged, contributing to nearly 5 million deaths annually worldwide from drug-resistant infections.
Traditional drug discovery methods, reliant on screening known chemical libraries, have struggled to keep pace.
But MIT’s Antibiotics-AI Project is flipping the script by using generative AI to explore uncharted “chemical spaces”, vast realms of hypothetical molecules that don’t exist in nature or labs yet.
Led by James Collins, the Termeer Professor of Medical Engineering and Science at MIT’s Institute for Medical Engineering and Science, the team generated over 36 million potential compounds computationally.
These were then screened using machine-learning models trained to predict antibacterial activity, toxicity to human cells, and novelty.
The result? Antibiotics that are structurally unlike any on the market, operating through fresh mechanisms that rupture bacterial cell membranes, making it harder for resistance to evolve.
“We’re excited about the new possibilities that this project opens up for antibiotics development,” Collins said in a statement.
The researchers employed two innovative strategies. For N. gonorrhoeae, they adopted a “fragment-based” approach.
Starting with a library of 45 million chemical fragments, building blocks of atoms like carbon, nitrogen, and oxygen, they used AI to identify promising ones with antimicrobial potential.
After filtering out toxic or familiar structures, they focused on a fragment dubbed F1. Generative algorithms, including one called chemically reasonable mutations (CReM) and another known as fragment-based variational autoencoder (F-VAE), then expanded F1 into full molecules.
From 7 million candidates, the team synthesized and tested two, with one of the candidates, NG1, proving highly effective.
In lab dishes and mouse models of drug-resistant gonorrhea, NG1 eradicated the bacteria by targeting LptA, a protein essential for building the bacterial outer membrane. This interference disrupts membrane synthesis, leading to cell death.
For MRSA, the team went unconstrained, letting AI algorithms freely invent molecules without a starting fragment. CReM and F-VAE churned out 29 million compounds, narrowed to 90 after screening
Of the 22 synthesized, six showed potent activity, with the standout DN1 clearing MRSA skin infections in mice. Unlike NG1, DN1’s membrane-disrupting effects are broader, not tied to a single protein.
This builds on prior MIT successes, like the AI-discovered antibiotics halicin and abaucin. Postdocs Aarti Krishnan and Melis Anahtar, along with recent PhD graduate Jacqueline Valeri, led the effort.
Now, nonprofit Phare Bio, part of the Antibiotics-AI Project, is refining NG1 and DN1 through medicinal chemistry, aiming for preclinical trials.
By venturing beyond known chemistry, MIT offers hope in the fight against superbugs, potentially saving millions of lives. As Collins noted, the approach addresses resistance “in a fundamentally different way,” paving the path for a new era of antibiotics.
