Researchers at the Department of Microbiology, Tumor and Cell Biology at Karolinska Institutet have co-authored a study published in Cell, which introduces a novel approach to antibiotic design using generative deep learning. The study, which is an international collaboration between Massachusetts Institute of Technology (MIT) and Karolinska Institutet, leverages deep learning to generate structurally novel antibiotics and proteomics to identify the cognate targets on a cell-wide scale.
Antibiotic resistance poses a growing threat to global health, with the discovery of few new antibiotic classes in recent decades. This study presents a generative deep learning platform for designing entirely new molecules with selective activity against pathogenic bacteria, such as Neisseria gonorrhoeae and Staphylococcus aureus.
"We are entering an era where generative AI is driving the discovery of entirely new antibiotics to combat some of the toughest bacterial pathogens - by exploring untapped regions of chemical space", says Prof. Collins, at Massachusetts Institute of Technology (MIT).
The team synthesized 24 AI-designed compounds, seven of which demonstrated selective antibacterial activity. Two lead molecules, NG1 and DN1, showed potent effects in mouse models of N. gonorrhoeae and methicillin-resistant S. aureus infections, with minimal toxicity and distinct mechanisms of action.
"Using two different generative AI approaches, we designed molecules that look nothing like known antibiotics or what pathogens would face in nature, overcoming the resistance crisis in a fundamentally different way", says Dr. Aarti Krishnan, lead author of the study at MIT.
Researchers from Karolinska Institutet identified the mechanism of action of the lead compound, NG1, a promising narrow-spectrum antibiotic designed de novo by AI. Using cutting-edge proteomics, they discovered that NG1 acts on LptA, an essential bacterial protein for transporting lipooligosaccharides, which form the outer membrane of N. gonorrhoeae, the causative agent of gonorrhea, thereby compromising the integrity of the pathogen's protective shield.
"Our proteomics-based analysis demonstrated that NG1 targets LptA. This discovery validates the power of combining AI-driven design with experimental target deconvolution," says Dr. Amir Ata Saei , co-author and Assistant Professor at the Department of Microbiology, Tumor and Cell Biology.
Looking ahead, the researchers believe their generative AI framework can be applied to other challenging pathogens.
This interdisciplinary effort underscores the importance of proteomics in target deconvolution and mechanistic validation, positioning Karolinska Institutet researchers to help pave the way for next-generation antibiotics.
Collaboration
The study was a collaboration between researchers at Karolinska Institutet, MIT, Harvard University and others.
Publication
" A generative deep learning approach to de novo antibiotic design "
Aarti Krishnan, Melis N. Anahtar, Jacqueline A. Valeri, Wengong Jin, Nina M. Donghia, Leif Sieben, Andreas Luttens, Yu Zhang, Seyed Majed Modaresi, Andrew Hennes, Jenna Fromer, Parijat Bandyopadhyay, Jonathan C. Chen, Danyal Rehman, Ronak Desai, Paige Edwards, Ryan S. Lach, Marie-Stéphanie Aschtgen, Margaux Gaborieau, Massimiliano Gaetani, Samantha G. Palace, Satotaka Omori, Lutete Khonde, Yurii S. Moroz, Bruce Blough, Chunyang Jin, Edmund Loh, Yonatan H. Grad, Amir Ata Saei, Connor W. Coley, Felix Wong, James J. Collins
Cell 2025 Aug.
