New profiling method speeds up tuberculosis drug discovery

Antibiotic resistance in tuberculosis is rising, and researchers urgently need better ways to discover and prioritize new drugs. Traditional whole-cell screens find compounds that kill or weaken Mycobacterium tuberculosis but usually do not reveal how those compounds work, making it hard to choose the most promising candidates for further development. To address that gap, a team led by corresponding author Deborah T. Hung expanded a strategy called PROSPECT ( PR imary screening O f S trains to P rioritize E xpanded C hemistry and T argets). PROSPECT measures chemical-genetic interactions by testing small molecules against a pooled collection of M. tuberculosis mutants, each depleted for a different essential protein target. This approach sensitively detects whole-cell activity while giving early clues about mechanism of action (MOA). Building on PROSPECT, the researchers developed a reference-based approach to interpret often complex interaction data so that MOA can be inferred sooner in the discovery pipeline. They curated a carefully annotated reference set of 437 compounds with published MOA and known or suspected antitubercular activity, and used these data to train a computational classifier aimed at converting PROSPECT fingerprints into MOA predictions.

The team created P erturbagen CL ass (PCL) analysis, a computational method that matches the chemical-genetic interaction profile of an unknown compound to profiles in the 437-compound reference set to predict MOA. They validated PCL analysis in a leave-one-out cross-validation, where each compound is predicted using the remaining reference compounds, and achieved 70% sensitivity and 75% precision. To test generalizability, they applied PCL analysis to 75 antitubercular leads with known MOA previously reported by GlaxoSmithKline (GSK), yielding 69% sensitivity and 87% precision. They then analyzed 98 GSK compounds that lacked MOA information and predicted that 60 of them act via a reference MOA; the team followed up with functional validation of 29 compounds predicted to target respiration-related MOAs. Finally, they applied PROSPECT and PCL analysis to approximately 5,000 compounds from larger unbiased libraries that had not been preselected for antitubercular activity. From that larger set, PCL analysis identified a novel scaffold that initially lacked wild-type activity but was predicted to inhibit respiration via QcrB; the prediction was confirmed and the scaffold was chemically optimized to achieve wild-type activity.

Together, these results show that combining PROSPECT with PCL analysis can turn complex chemical-genetic data into actionable MOA predictions, allowing researchers to prioritize hits based on likely biological targets rather than just cell killing. The approach works both for compound sets already suspected to affect M. tuberculosis and for large, unbiased libraries, increasing the chances of finding novel scaffolds that might otherwise be missed. Importantly, the method can flag compounds that do not yet show activity against wild-type bacteria but appear to hit a validated target such as QcrB, providing a clear path for chemical optimization. By delivering MOA information early, the strategy reduces wasted effort on compounds without clear targets and focuses follow-up studies on leads with higher chances of development. In an era of growing resistance, the ability to assign MOA quickly and at scale could help accelerate the discovery of new, potent antitubercular compounds and guide resource allocation during drug development.