Repurposing Impact Project
Development of a scoring system that integrates predicted likelihood of success, burden of illness, and economic impact for drug-disease pairs to facilitate selection of high-impact opportunities for further advancement
Supported by Arnold Ventures
Introduction
A unique opportunity exists to find new uses for drugs already approved by the Food & Drug Administration (FDA). This process, called drug repurposing, could help treat other disease(s) that currently lack good treatment options. The challenge is how to select from the approved drugs to pair with the many potential diseases needing treatments, i.e. potential drug-disease pairings. To address this challenge, we are developing a novel scoring system (algorithm) to systematically identify promising drugs to pair with diseases to maximize impact on patients and decrease the economic impact on the health care system. Our approach integrates large language models (LLMs), rule-based artificial intelligence (AI), and informatics, to quickly analyze vast amounts of data. This approach prioritizes the most promising candidates for further investigation, ensuring that our partner, Every Cure, and other research groups, can focus their resources on the most promising therapeutic interventions, ultimately leading to improved patient outcomes and reducing healthcare costs.
Initial Results
The current version of the algorithm includes three key factors when scoring drug-disease pairs: 1) patient impact component score, 2) biological plausibility, which evaluates the likelihood that the drug would be effective, and 3) economic impact score, which evaluates the potential cost-effectiveness of repurposing the drug. The final results of the model are the individual component scores and the final prioritization score as a weighted average of all component scores. Currently, we weight all components equally, but we will determine appropriate weights through expert-based evaluation over the next few months.
To date, the algorithm was run on approximately 50,000 drug-disease pairs out of a possible 5.7 million. The list of the top 100 ranked disease-drug pairs, can be downloaded below. Each row in the table provides the names of the drug and disease, standardized codes for drugs and diseases, and a weighted average of the individual scores that comprises the final prioritization score for the drug-disease pair.
We plan to refine the algorithm’s outputs, add additional data sources, perform evaluation to assess the quality of the algorithm’s outputs, and develop an interactive visualization dashboard with improved tools to browse and download the resulting data. We plan to run the refined algorithm on a broader list of drug-disease pairs to determine potential drug repurposing opportunities for further investigation as treatment options.