In the face of extensively drug-resistant superbugs, there's a pressing need for novel antibiotics. Recently, researchers from McMaster University in Canada and Stanford University in the United States have developed a new generative artificial intelligence model capable of swiftly and affordably designing new antibiotic molecules and providing synthetic routes, enabling chemists to easily synthesize these molecules in the lab. On March 22, their findings were published in Nature Machine Intelligence.
A global study in 2022 revealed that around 1.27 million people worldwide died directly from antibiotic resistance in 2019.
"Antibiotics are a unique class of drugs. Once we start using them clinically, we're essentially starting a countdown to drug failure as bacteria rapidly evolve to resist them," remarked Jonathan Stokes, assistant professor at McMaster University and lead author of the paper. "We need to discover antibiotics quickly and inexpensively, and that's where artificial intelligence plays a crucial role."
Previously, AI models faced significant limitations in antibiotic development: property prediction models required evaluating different molecules one by one, with poor scalability, while generative models, although capable of rapidly designing multiple molecules, often generated molecules that were challenging to synthesize in the lab.
The researchers devised an AI generative model called SyntheMol and tested it using the extensively drug-resistant bacterium Acinetobacter baumannii, recognized by the World Health Organization as one of the most dangerous drug-resistant bacteria globally.
They extracted data from a database containing 132,000 molecular fragments, which could seamlessly connect like LEGO bricks but possessed vastly different properties. These molecular fragments were then cross-validated with a set of 13 chemical reactions, resulting in bidirectional combinations of 30 billion fragments. From these, they designed 58 promising new molecules and used another AI toxicity prediction model to assess their toxicity.
In subsequent lab experiments, they synthesized these structurally novel molecules and validated that six of them exhibited effective antibacterial activity against Acinetobacter baumannii and several other bacterial pathogens, without toxicity.
"SyntheMol not only designs novel molecules for promising drug candidates but also generates the recipes for making each new molecule. This is a new approach and a game-changer because chemists often don't know how to make AI-designed molecules," said James Zou, co-author of the paper and associate professor at Stanford University.
For more information on the paper, visit: https://doi.org/10.1038/s42256-024-00809-7
