Searching for new treatments for Alzheimer’s disease (AD), researchers at Vanderbilt University Medical Center asked ChatGPT whether any existing drugs used for other problems might help patients with AD.
ChatGPT (Chat Generative Pre-trained Transformer) is a cloud-based artificial intelligence (AI) program, a so-called large language model (LLM) that’s been optimized for chat. Created by San Francisco based OpenAI, Inc., the chatbot has quickly attracted an estimated 180 million users.
The team prompted the chatbot to review the biomedical research literature and suggest 20 drugs for AD, ranking them for potential effectiveness. With a second prompt they asked for confirmation of the initial 20 suggestions. They repeated both prompts 10 times.
Proceeding with the bot’s 10 most frequently suggested drugs, the team analyzed medical records of patients age 65 and older from VUMC and the All of Us Research Program to see whether those exposed to any of the 10 drugs had lower rates of AD.
In a combined analysis across datasets from the two organizations, after 10 years of follow-up three of the 10 drugs were found to be associated with lower risk of AD.
- Losartan, used to treat high blood pressure and heart failure, was associated with 24% reduced risk of AD.
- Metformin, used to treat Type 2 diabetes, was associated with 33% reduced risk of AD.
- Simvastatin, used to treat high cholesterol and fat levels in the blood, was associated with 16% reduced risk of AD.
In the VUMC dataset, minocycline, an antibiotic, was associated with 66% reduced risk of AD; an association with minocycline could not be measured in All of Us because too few of its research subjects with AD were exposed to the drug.
The authors note that preclinical and clinical testing of drug repurposing candidates is ongoing and has not yet provided grounds for repurposing any drugs to help AD patients.
The research team reported the study Feb. 26 in Nature Digital Medicine. The study was led by Chao Yan, PhD, a postdoctoral research fellow in Biomedical Informatics, Monika Grabowska, an MD-PhD student, and Wei-Qi Wei, MD, PhD, associate professor of Biomedical Informatics.
“LLMs like ChatGPT speedily accomplish a form of extensive literature review, which has become infeasible for humans to perform alone. Our research indicates that it produces high-quality hypotheses for repurposing drugs,” said Wei, the paper’s corresponding author and one of its two senior authors with Bradley Malin, PhD, professor of Biomedical Informatics.
“Pipelines that leverage the capabilities of LLMs like ChatGPT would appear to offer a streamlined new framework for drug repurposing that can be applied to numerous diseases,” Wei said.
Others on the study from Vanderbilt include Alyson Dickson, MA, Bingshan Li, PhD, Dan Roden, MD, C. Michael Stein, MD, Peter Embí, MD, MS, Josh Peterson, MD, MPH, and QiPing Feng, PhD. The study was supported by the National Institutes of Health (R01GM139891, R35GM131770, R01AG069900, F30AG080885).