Machine Learning-Driven Small Molecule Design of Klebsiella-Specific Antibiotics

Yves Brun of the University of Montréal and Mike Tyers of the Hospital for Sick Children in collaboration with colleagues at the Québec Institute for Learning Algorithms, the Institute for Research in Immunology and Cancer, SickKids, the University of Toronto and Simmunome Inc., all in Canada, will combine generative machine learning (ML) with high-throughput phenotypic- and target-based screens to identify new antibiotics against multidrug-resistant Klebsiella. The multidisciplinary team will use high-content microscopy and genome-wide CRISPRi to generate phenotypic and genetic profiles of Klebsiella responses to compounds and then train ML models on antibiotic activity, penetration, and resistance. In parallel, the team will use generative ML to design novel, synthesizable compounds against key Klebsiella targets, which will be produced by parallelized chemical synthesis and tested for antibiotic activity. A lab-in-the-loop active learning approach will be used to iteratively optimize ML models to predict potent new antibiotics active against Klebsiella.