Overcoming Multidrug-Resistance in Bacteria with a Two-Step Process to Repurpose and Recombine Established Drugs.

Abstract

The emergence and ongoing spread of multidrug-resistant (MDR) bacteria is a major global public health threat. MDR has extensively combated the potency of antibiotics. Development of new antibiotics requires several years with prohibitive cost that will not last. An alternative solution is to recombine failed antibiotics, which has been proven to be not only cost-effective, but also potent. However, selection of the optimal combinations of these chemicals through conventional trial-and-error methods is challenging and slow, since M candidates with N doses lead to NM possible combinations. Herein, we present a artificial intelligence (AI) guided chemical combination optimization technique, namely Streamlined Rapid Identification of Combinatorial Therapies (STRICT), which is phenotype based and can efficiently learn and identify the optimal drug-combinations with minimal experimental efforts. With the guidance of STRICT, we successfully identified potent combinations of five antibiotics from 26 antibiotics that are individually ineffective at inhibiting an artificially induced strain of MDR bacteria. Rather than examine millions of tests, STRICT accomplished this task with only 120 carefully selected tests. Our results indicate that STRICT is a powerful platform to identify efficacious multiantibiotic combinations for the treatment of MDR bacteria. The AI-guided platform introduced here is an effective tool for drug repurposing, beneficial toward large-scale drug screening for other disease models, and also has a broad application in chemical combination optimization to deliver a desired end point for a complex system.