MERIT: Mechanism driven model predicts drug outcomes and nominates indications for failed drugs

In the realm of drug development, efficacy and safety are paramount concerns. However, many existing trial-outcome prediction models rely on trial design, prior development history, or compound identity, leading to compound memorization and inflated performance metrics. A novel model named Mechanism-Resolved Inference of Trial outcomes (MERIT) has been developed to address these shortcomings.

MERIT relies solely on molecular and disease features to predict trial outcomes, without incorporating information on the success of similar compounds.

This innovative approach integrates large-scale drug-protein, protein-metabolite, and immune interaction maps to establish connections between a drug's intended effects and its potential off-target effects. By analyzing tissue-specific efficacy and safety, MERIT provides a comprehensive understanding of a drug's potential performance.

When tested across 753 small-molecule drugs and 3,133 trials, MERIT demonstrated exceptional performance, achieving the highest overall AUROC (Area Under the Receiver Operating Characteristic curve) value of 0.770. This figure indicates that MERIT excelled in predicting both efficacy (0.765) and safety (0.784) outcomes. Furthermore, MERIT was able to recover the eventual approved indications for 83% of failed drugs, showcasing its ability to uncover potential new uses for underperforming compounds.

Beyond retrospective analysis, MERIT also played a proactive role in ongoing research by registering locked, outcome-blind predictions for 55 drug-indication pairs in ongoing Phase III trials. This unique feature establishes MERIT as a prospective evaluation cohort, providing a valuable resource for researchers and clinicians involved in the drug development process.