Prédiction de la résistance aux antibiotiques à partir de données génomiques à l'aide d'approches d'apprentissage automatique et d'apprentissage profond

Antibiotic resistance is a worldwide health issue spreading quickly among human and animal pathogens, as well as environmental bacteria. Recent developments in whole genome sequencing and in machine learning models provide important opportunities that are partially exploited in antibiotic resistance. This is in part due to the limited availability of input data required to train the models. Hence, as part of the Seq2Diag project, we collected genomic data of three clinically relevant bacteria (Escherichia coli, Klebsiella pneumoniae, Pseudomonas aeruginosa) and the associated resistance profile for several antibiotics. Based on those data, we developed machine learning models and genomic language models for in silico prediction of antibiotic resistance profiles using the DNA sequencing data. After validation of the models, these predictions may be used in real-time to guide the personalized treatment of patients in clinical settings, as an alternative to phenotypic characterization of antibiotic resistance.