AI designs novel antibiotics to combat drug-resistant superbugs

MIT researchers have leveraged generative artificial intelligence to design a new class of antibiotics, including compounds that show significant promise against two notoriously difficult-to-treat infections: drug-resistant gonorrhea and multi-drug-resistant Staphylococcus aureus (MRSA). This groundbreaking work, detailed in the journal Cell, represents a significant leap in the fight against a global health crisis, demonstrating AI’s power to explore vast, previously inaccessible chemical landscapes in the quest for novel therapeutics.

The urgency for new antibiotics has never been greater. Over the past 45 years, few truly novel antibiotics have gained FDA approval; most are variations of existing drugs. Meanwhile, bacterial resistance continues its alarming rise, contributing to an estimated 5 million deaths annually worldwide. To counter this, Professor James Collins, a senior author of the study and a leader in MIT’s Antibiotics-AI Project, and his team expanded their previous AI-driven screening efforts, which had already yielded promising candidates like halicin and abaucin. This time, their focus shifted from screening existing compounds to designing entirely new ones.

Using generative AI algorithms, the research team computationally designed over 36 million hypothetical compounds and then screened them for antimicrobial properties. The top candidates identified are structurally distinct from any known antibiotics, suggesting they operate through novel mechanisms, primarily by disrupting bacterial cell membranes. “We’re excited about the new possibilities that this project opens up for antibiotics development,” Collins stated. “Our work shows the power of AI from a drug design standpoint, and enables us to exploit much larger chemical spaces that were previously inaccessible.”

The researchers pursued two distinct AI-driven strategies. In the first approach, they directed generative AI algorithms to design molecules based on a specific chemical fragment known to possess antimicrobial activity. Targeting Neisseria gonorrhoeae, the bacterium responsible for gonorrhea, they began by sifting through a library of approximately 45 million known chemical fragments. Through multiple computational screening rounds, they identified a promising fragment, dubbed F1, which showed activity against N. gonorrhoeae while avoiding cytotoxicity to human cells or similarities to existing antibiotics. This deliberate avoidance of known structures aimed to circumvent existing resistance mechanisms.

Using F1 as a foundation, two different generative AI algorithms—chemically reasonable mutations (CReM) and fragment-based variational autoencoder (F-VAE)—generated about 7 million new candidate compounds. Subsequent computational screening narrowed this down to roughly 1,000 compounds. Of the 80 selected for potential synthesis, only two proved chemically feasible to produce. One of these, named NG1, demonstrated remarkable efficacy against N. gonorrhoeae in both laboratory dishes and a mouse model of drug-resistant gonorrhea. Further investigation revealed that NG1 targets a protein called LptA, which is crucial for the synthesis of the bacterial outer membrane, effectively disrupting this vital process.

For their second approach, the team explored the potential of unconstrained generative AI design, targeting the Gram-positive bacterium Staphylococcus aureus. Here, the CReM and F-VAE algorithms were given greater freedom, generating over 29 million compounds based only on general rules of chemical plausibility. After applying similar filtering criteria, the pool was reduced to about 90 candidates. Twenty-two of these were synthesized and tested, with six exhibiting potent antibacterial activity against multi-drug-resistant S. aureus in lab settings. The leading candidate, DN1, successfully cleared a methicillin-resistant S. aureus (MRSA) skin infection in a mouse model. Like NG1, these molecules also appear to interfere with bacterial cell membranes, though with broader effects not limited to a single protein target.

The nonprofit Phare Bio, a partner in the Antibiotics-AI Project, is now working to further modify NG1 and DN1 for advanced preclinical testing. Collins expressed optimism about applying these platforms to combat other critical bacterial pathogens, including Mycobacterium tuberculosis and Pseudomonas aeruginosa. This pioneering research underscores AI’s transformative potential in drug discovery, offering a beacon of hope in the escalating battle against antimicrobial resistance.