Amid a growing global health crisis fueled by antibiotic-resistant “superbugs,” researchers at MIT have harnessed generative AI to propel antibiotic innovation beyond traditional discovery methods, reports IEEE Spectrum. In a study published in Cell, led by Professor Jim Collins of MIT’s Biological Engineering department, the team trained AI systems on extensive antibacterial datasets to conceive millions of entirely novel molecular structures—molecules that don’t exist in any chemical library.
These generative algorithms, fueled by rich neural network frameworks, began by virtually screening over 45 million chemical fragments. The AI models sifted through these data to identify promising building blocks with potential activity against Neisseria gonorrhoeae and Staphylococcus aureus, including multi-drug-resistant strains such as MRSA. Rather than searching for hidden compounds in known libraries, the AI designed de novo molecules from scratch—completely new potential antibiotics.
From the AI-generated pool, the researchers synthesized a select number of molecules and tested them in laboratory assays and mouse models. Two standout candidates, NG1 (targeting gonorrhea) and DN1 (effective against MRSA), demonstrated potent antibacterial activity and succeeded in clearing infections in vivo. Notably, NG1 appears to disrupt bacterial survival by binding to a protein involved in outer-membrane synthesis, while DN1 broadly compromised bacterial cell membranes.
This marks a pivotal shift: AI has evolved from a tool that discovers compounds within existing chemical space to one that constructs entirely new molecular entities—a profound leap in antibiotic research. The research, described as “elegant and potentially clinically meaningful” by experts in the field, lays groundwork for exploring uncharted chemical territories in the fight against antimicrobial resistance.
Looking ahead, Phare Bio—an MIT-affiliated social enterprise—plans to advance these AI-conceived antibiotics through pre-clinical and clinical development stages.