Design principles to assemble drug combinations for effective tuberculosis therapy using interpretable pairwise drug response measurements.

Abstract

A challenge in tuberculosis treatment regimen design is the necessity to combine three or more antibiotics. We narrow the prohibitively large search space by breaking down high-order drug combinations into drug pair units. Using pairwise in vitro measurements, we train machine learning models to predict higher-order combination treatment outcomes in the relapsing BALB/c mouse model. Classifiers perform well and predict many of the >500 possible combinations among 12 antibiotics to be improved over bedaquiline + pretomanid + linezolid, a treatment-shortening regimen compared with the standard of care in mice. We reformulate classifiers as simple rulesets to reveal guiding principles of constructing combination therapies for both preclinical and clinical outcomes. One example ruleset combines a drug pair that is synergistic in a dormancy model with a pair that is potent in a cholesterol-rich growth environment. These rulesets are predictive, intuitive, and practical, thus enabling rational construction of drug combinations.