Viagra (sildenafil) was not originally intended to help with erectile dysfunction. That outcome was a serendipitous side effect of the drug, which was developed to treat high blood pressure and angina. But drug makers at Pfizer rebilled the drug for its other uses. Such repurposing of drugs that are already well-characterized and safe for use can speed up the drug discovery process.
Now, researchers leverage the power of artificial intelligence and machine learning to mine real-world data from electronic health records, insurance claims, and patient surveys to purposefully identify drug candidates for repurposing.
“Randomized clinical trials represent the strongest evidence for drug discovery,” Ping Zhang, a computer scientist at The Ohio State University in Columbus, Ohio who led the new research, said via email. But, he added, “They are costly, slow, and often impractical to generate evidence for many important questions.”
To overcome these problems, Zhang and colleagues built a customizable framework that emulates randomized clinical trials. For a given disease, the framework uses data from electronic health records and medical claims to extract a list of drugs or active ingredients with potential for repurposing. Then it identifies users and nonusers for each ingredient and uses a propensity score estimation model based on deep learning to account for confounding biases, such as whether a patient is on more than one medication, and estimate the effect of the drug or ingredient, setting the stage for drug repurposing.
Using the framework, Zhang and team successfully identified six coronary artery disease (CAD) drug candidates that have beneficial effects on disease outcomes, but are not yet indicated for the heart condition. Among the potential repurposing candidates were metformin, a first-line treatment for diabetes, and escitalopram, prescribed for major depressive disorder or generalized anxiety, the researchers reported in Nature Machine Intelligence.
The framework also assessed drug combinations for CAD. “Interestingly, lisinopril and atorvastatin are not statistically significant as individual treatments for CAD,” Zhang said. “But their combination, i.e., taking them together, substantially improves CAD disease outcomes.”
When compared against three existing preclinical drug repurposing methods, the new approach outperformed them all, suggesting that it has legs.
“It’s a nice framework,” said Hongfang Liu, a computer scientist at the Mayo Clinic in Rochester, Minnesota, who does similar research, but was not part of the new study. “It highlights the potential opportunity to use real world data for simulating potential candidates’ effects.”
Plus, it targets one challenge of using randomized clinical trials to screen for repurposing drug candidates: cost. “It’s a much, much cheaper way,” said Liu.
For Zhang, the real power of the framework is in its hypothesis generation. “I want to find new uses for old drugs for diseases…like Parkinson’s disease, Alzheimer’s disease, and Huntington’s,” he said. “This method can be a very good resource to study these diseases as well.”
“The findings are hypotheses generated by artificial intelligence [using] real-world data, which pave the way for drug repurposing,” he said.
- Liu, R., Wei, L. & Zhang, P. A deep learning framework for drug repurposing via emulating clinical trials on real-world patient data. Nat Mach Intell 3, 68–75 (2021).