AI and human scientists work together to discover new cancer drug combinations

AI and human scientists work together to discover new cancer drug combinations

An "AI scientist" working in collaboration with human scientists has found that combinations of cheap and safe drugs - used to treat conditions such as high cholesterol and alcohol addiction - could also be effective in treating cancer, a promising new approach to drug discovery.

The University of Cambridge-led research team used the GPT-4 large language model (LLM) to identify hidden patterns buried in the mountains of scientific literature to identify potential new cancer drugs.

To test their approach, researchers led GPT-4 to identify potential new drug combinations that could have a significant impact on a breast cancer cell line commonly used in medical research. They directed it to avoid standard cancer drugs, identify drugs that attack cancer cells without harming healthy cells, and prioritize drugs that were affordable and approved by regulators.

GPT-4's proposed drug combinations were then tested by human scientists both in combination and individually to measure their effectiveness against breast cancer cells.

In the first laboratory-based test, three of GPT-4's 12 proposed drug combinations worked better than current breast cancer drugs. The LLM then learned from these tests and proposed a further four combinations, three of which also showed promising results.

The results, reported in theJournal of the Royal Society Interfacerepresent the first instance of a closed-loop system in which experimental results guided an LLM, and LLM outputs-interpreted by human scientists-guided further experiments. The researchers say that tools like LLMs will not replace scientists, but could instead supervise AI researchers, with the ability to emerge, adapt and accelerate discovery in areas such as cancer research.

Often LLMs like GPT-4 return results that are not true are known as hallucinations. However, in scientific research, hallucinations can sometimes be an advantage if they lead to new ideas worth testing.

“Supervisory LLMs provide a scalable, imaginative layer of scientific exploration and can help us as human scientists investigate new avenues that we had not previously thought of,” said Professor Ross King from the Department of Chemical Engineering and Biotechnology at Cambridge, who led the research. “This can be useful in areas such as drug discovery, where many thousands of compounds need to be screened.”

Based on input from human scientists, GPT-4 selected drugs based on the interplay between biological thinking and hidden patterns in the scientific literature.

This is not automation replacing scientists, but a new way of collaborating. Guided by expert prompts and experimental feedback, the AI ​​acted like a tireless research partner, navigating an immense hypothesis space and suggesting ideas that would take humans longer to achieve on their own. “

Dr. Hector Zenil, co-author from King's College London

The hallucinations – normally considered defects – became a feature, creating unconventional combinations worth testing and validating in the laboratory. The human scientists examined the mechanistic reasons why the LLM suggested these combinations in the first place, feeding the system back and forth in multiple iterations.

By exploring subtle synergies and overlooked pathways, GPT-4 helped identify six promising drug pairs, all tested through laboratory experiments. Among the combinations, simvastatin (commonly used to lower cholesterol) and disulfiram (used in alcohol addiction) stood out against breast cancer cells. Some of these combinations show the potential for further research in therapeutic reparation.

These drugs, while not traditionally associated with cancer care, could be potential cancer treatments, although they would first need to conduct large-scale clinical trials.

“This study shows how AI can be integrated directly into the iterative loop of scientific discovery, enabling the generation and validation of the data-informed hypotheses in real time,” said Zenil.

“The capacity of supervised LLMs to propose hypotheses across disciplines, consider prior findings, and collaborate on iterations across a new frontier in scientific research,” King said. “An AI scientist is no longer a metaphor without experimental validation: it can now be a collaborator in the scientific process.”

The research was supported in part by the Alice Wallenberg Foundation and the UK Engineering and Physical Sciences Research Council (EPSRC).

Sources: