Data-driven Inference of Drug Synergy
Combination therapy is an important strategy for treating tuberculosis (TB), which kills almost 2 million people each year. Long treatment durations and growing rates of multi-drug resistant (MDR) TB have underscored the need for entirely new multi-drug regimens rather than single agents to combat the TB pandemic. Because empirically assessing the astronomically immense set of all possible drug combinations is prohibitive, there is pressing need for rational approaches to prioritize new drug regimens for clinical trials. To begin addressing this problem, we adapted a computational tool that we recently created—Inferring Drug Interactions using chemo-Genomics and Orthology, (INDIGO)—which accurately predicts drug synergy/antagonism in E. coli based on the available high resolution chemogenomic data available for that organism. We modified INDIGO to predict synergy/antagonism in 26,106 2-way and over 1 million 3-way interventions involving 164 compounds and 65 perturbations in Mycobacterium tuberculosis (MTB), by leveraging publicly available transcriptomic data. In vitro validation of predicted 2-way and 3-way interactions by checkerboard assay revealed strong correlation with model predictions, and model predictions also correlated significantly with 2-month sputum conversion rates elicited by multidrug regimens as reported from clinical trials. In addition, INDIGO analysis has revealed approximately 250 highly conserved genes that are most predictive of drug interaction outcome. We have integrated this core gene set and MTB network information to identify pathways and regulators that influence drug interaction outcomes, which in turn reveals new targets for combination therapy. INDIGO shows great promise for efficiently selecting novel TB drug regimens by prioritizing combinations based on extent of synergy or antagonism and by identifying mediators that can enhance these interactions.