AI helps scientists uncover potential cancer-fighting use for cholesterol and alcohol dependence drugs
An ‘AI scientist’ working alongside human researchers has helped identify combinations of affordable, approved drugs—originally used for conditions like high cholesterol and alcohol dependence—that may also be effective against cancer.
The study, led by researchers at the University of Cambridge, used the GPT-4 large language model (LLM) to find hidden patterns in scientific literature and suggest potential cancer treatments.
GPT-4 was specifically prompted to propose drug combinations that could target breast cancer cells without harming healthy tissue.
The researchers asked it to avoid conventional cancer drugs and prioritise low-cost, regulator-approved options.
The AI suggested 12 combinations, which were then tested in the lab against a breast cancer cell line.
Three of the combinations outperformed current breast cancer treatments. Based on the lab results, the AI generated four further combinations, three of which also showed promising effects.
This marks the first instance of a closed-loop system in which experimental results informed the LLM, which then proposed new experiments—creating a continuous feedback cycle between AI and scientists.
Among the standout combinations were simvastatin, commonly used to lower cholesterol, and disulfiram, used to treat alcohol dependence.
While these drugs are not typically linked to cancer care, they demonstrated effectiveness against breast cancer cells in laboratory experiments.
Professor Ross King from Cambridge’s Department of Chemical Engineering and Biotechnology, led the research.
He said: “Supervised LLMs offer a scalable, imaginative layer of scientific exploration, and can help us as human scientists explore new paths that we hadn’t thought of before.
“This can be useful in areas such as drug discovery, where there are many thousands of compounds to search through.”
The researchers noted that large language models like GPT-4 can produce errors, known as hallucinations.
However, in this context, such hallucinations occasionally led to creative and unconventional ideas worth testing further.
“This is not automation replacing scientists, but a new kind of collaboration,” said Dr Hector Zenil, co-author from King’s College London.
“Guided by expert prompts and experimental feedback, the AI functioned like a tireless research partner—rapidly navigating an immense hypothesis space and proposing ideas that would take humans alone far longer to reach.”
The AI chose drugs by identifying biological reasoning and overlooked links in published research—insights that human scientists may have missed.
Through several cycles of AI suggestions and laboratory testing, six promising drug combinations were ultimately identified.
While the findings point to new possibilities for repurposing existing medications, the researchers emphasised that all combinations would need to undergo extensive clinical trials before being used as cancer treatments.
