An ‘AI scientist’, working in collaboration with human scientists, has discovered that combinations of cheap and safe drugs could be effective at treating breast cancer
The researchers, led by the University of Cambridge, utilised the GPT-4 large language model (LLM) to identify hidden patterns within extensive scientific literature, aiming to discover potential new cancer drugs.
The results are detailed in the Journal of the Royal Society Interface.
The drug combinations were sourced by AI but tested by human scientists
The researchers promoted GPT-4 to identify potential new drug combinations that could have a significant impact on breast cancer cell lines commonly used in medical research.
They instructed the AI software to avoid standard cancer drugs, identify drugs that would attack cancer cells without harming healthy cells, and prioritise drugs that were affordable and approved by regulators.
GPT-4 suggested various drug combinations that were tested by human scientists, both in combination and individually, to measure their effectiveness against breast cancer cells.
Three of the 12 drug combinations worked better than the current treatment
The first lab test revealed that three of the 12 drug combinations suggested by AI worked better than current breast cancer drugs. The tool learned from these tests and suggested four additional combinations, three of which showed promising results.
The results represent the first instance of a closed-loop system where experimental results guided an LLM, and the LLM’s outputs, interpreted by human scientists, guided further experiments.
The researchers note that AI tools are not a replacement for scientists but could be utilised to help originate, adapt and accelerate drug and research discovery. Human scientists play a crucial role in guiding the AI, interpreting its outputs, and conducting the necessary experiments.
“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,” said Professor Ross King from Cambridge’s Department of Chemical Engineering and Biotechnology, who led the research. “This can be useful in areas such as drug discovery, where there are many thousands of compounds to search through.”
Based on the prompts provided by human scientists, GPT-4 selected drugs based on biological reasoning and hidden patterns it uncovered in scientific literature.
“This is not automation replacing scientists, but a new kind of collaboration,” said co-author Dr Hector Zenil 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.”
In some cases, LLMs can return results that aren’t true, known as hallucinations. However, hallucinations can sometimes be beneficial if they lead to ideas worth investigating. The scientists explored the mechanistic reasons the AI tool suggested these combinations in the first place, providing feedback in multiple iterations.
By exploring subtle synergies and overlooked pathways, GPT-4 helped identify six promising drug pairs, all of which were subsequently tested through laboratory experiments. The combination of simvastatin (commonly used to lower cholesterol) and disulfiram (used in alcohol dependence) showed promise in reducing breast cancer cells. Some of these combinations show potential for further research in therapeutic repurposing.
“This study demonstrates how AI can be woven directly into the iterative loop of scientific discovery, enabling adaptive, data-informed hypothesis generation and validation in real-time,” said Zenil.
“The capacity of supervised LLMs to propose hypotheses across disciplines, incorporate prior results, and collaborate across iterations marks a new frontier in scientific research,” said King. “An AI scientist is no longer a metaphor without experimental validation: it can now be a collaborator in the scientific process.”
