SDSC: Using NSF ACCESS Supercomputers to Improve Tuberculosis Treatment Options

Key Takeaways

• Simulated 219 TB drug combos using NSF ACCESS supercomputers
• Machine learning model predicts outcomes, reducing need for exhaustive trials
• Identified regimens with shorter treatment and lower drug doses
• Computations consumed over 600,000 CPU hours across four clusters
• Approach could accelerate TB drug development in low‑resource settings

Pulse Analysis

Tuberculosis remains a leading cause of infectious mortality, with standard therapy lasting six to nine months and requiring a precise cocktail of antibiotics. The lengthy regimen drives poor adherence, especially in low‑income regions where drug supply chains are fragile. Researchers have therefore turned to high‑performance computing to model the disease at the cellular level, seeking shortcuts that can reveal more effective, patient‑friendly treatment strategies without the time and cost of traditional laboratory trials. Moreover, the rise of multidrug‑resistant TB strains intensifies the urgency for novel therapeutic designs.

The University of Michigan team leveraged NSF ACCESS allocations on the Expanse and Anvil supercomputers to run over 600,000 CPU hours of simulations, evaluating 219 distinct drug combinations against virtual granulomas in the lung. By first conducting a modest set of experiments and training a machine‑learning surrogate, they could predict the performance of untested regimens, dramatically shrinking the computational search space. This hybrid approach blends mechanistic modeling of host‑pathogen interactions with data‑driven inference, delivering a rapid ranking of candidates that balance efficacy, treatment duration, and total drug burden.

The identified regimens promise to cut treatment time and lower drug exposure, which could improve adherence and reduce side effects—critical factors for TB control in resource‑constrained settings. By demonstrating that national supercomputing resources can accelerate biomedical discovery, the study underscores the strategic value of NSF’s ACCESS program for public‑health challenges. Future work will likely integrate patient‑specific data and expand to other infectious diseases, turning high‑performance computing into a routine partner for precision medicine and global health initiatives. If validated clinically, these computationally derived protocols could reshape WHO treatment guidelines within the next decade.