Pilot study tests ability of generative AI to prioritize drugs for potential repurposing in Alzheimer’s disease.

Drug repurposing has emerged as a compelling alternative to traditional drug-development channels. Now, advancements in artificial intelligence have led researchers to explore AI-driven ways for more quickly discerning which drugs are the best candidates.

Wei-Qi Wei, M.D., Ph.D., an associate professor of biomedical informatics at Vanderbilt University Medical Center, said the work is especially important for patients who have prevalent conditions with few treatment options, such as Alzheimer’s disease.

The success of repurposing relies on rapid and accurate identification of promising candidates from a large pool of potential drugs, Wei explained. Among the models being investigated for this process are generative pre-trained transformers, or GPTs.

“GPTs and other large language models have the ability to interpret and respond to a wide range of inquiries,” Wei said. “These tools can be leveraged to accelerate review of the scientific literature, helping to prioritize drug repurposing candidates.”

LLMs as a Repurposing Tool

Wei led a pilot study recently reported in npj Digital Medicine that evaluated the feasibility of using ChatGPT to identify the most promising drugs for repurposing in Alzheimer’s. Using two sequential prompts, the researchers asked ChatGPT to generate names of the 20 most promising drugs to consider.

Wei and his team then looked for Alzheimer’s incidence in patients over 65 who had been prescribed one of the top 10 drug candidates identified by ChatGPT. The data came from two large clinical datasets: one from VUMC and another from the All of Us Research Program.

“When I first saw the list of candidates, I was shocked. The list was surprisingly rational, with some of the drugs already being studied for potential use in treating Alzheimer’s.”

Drug candidates in ChatGPT’s recommendations included metformin, losartan and simvastatin, all of which were associated with a lower risk of Alzheimer’s in meta-analysis.

“When I first saw the list of candidates, I was shocked,” Wei said. “The list was surprisingly rational, with some of the drugs already being studied for potential use in treating Alzheimer’s.”

According to Wei, ChatGPT’s ability to serve as a repurposing tool lies in its ability to quickly review and synthesize information from relevant literature. Notably, it did not recommend any FDA-approved drugs for Alzheimer’s, implying that the model could interpret the concept of drug repurposing.

ChatGPT is feasible for use as an AI-driven hypothesis generator for drug repurposing, Wei determined, enabling prompt generation of a promising list of drugs for subsequent testing in EHRs.

Models’ Pros and Cons

Wei noted that large language models like ChatGPT can greatly speed up the review process for long lists of drug candidates, giving researchers more time to test and validate their hypotheses.

Additionally, combining LLM-based findings with real-world clinical datasets could represent a cost-effective strategy to explore preliminary signals before allocating further resources to research and clinical trials, he said.

Despite these advantages, the tools are not without their drawbacks. Specifically, LLMs run the risk of overlooking promising drug candidates not frequently cited in literature. Furthermore, the reliability of LLM-generated responses can be affected by updates in data, learning methods and other metrics used in model training.

To overcome these limitations, Wei recommended ongoing monitoring of LLM performance, coupled with validation using larger clinical datasets.

Future Investigations

As research in this area evolves, Wei suggested that future studies should determine how best to leverage LLMs to uncover novel insights in combination with knowledge representation and reasoning technologies, such as knowledge graphs.

In drug development contexts, pipelines that apply the capabilities of LLMs offer a new framework for drug repurposing that can be applied to multiple indications, he added.

“I’m excited about the potential for future applications as we better understand the role of AI in drug discovery,” Wei said.

About the Expert

Wei-Qi Wei, M.D.

Wei-Qi Wei, M.D., Ph.D., is an associate professor of biomedical informatics at Vanderbilt University Medical Center. His research focuses on deep phenotyping through terminology/ontology, natural language processing, machine learning, and harnessing big data from EHR to make pharmacogenomics discoveries.