The AI's ability to generate unconventional combinations, sometimes referred to as 'hallucinations', has led to valuable insights and hypotheses that merit further investigation in the lab.

A research team from the University of Cambridge has harnessed the capabilities of the GPT-4 large language model to identify potential new combinations of existing non-cancer drugs that may effectively target and kill cancer cells.

The team specifically instructed the AI to focus on affordable, regulator-approved medications, deliberately avoiding standard cancer treatments.

Among the promising combinations identified were simvastatin, a cholesterol-lowering drug, and disulfiram, used for alcohol dependence, both of which showed significant potential in killing breast cancer cells.

In initial lab tests, three out of twelve combinations recommended by GPT-4 outperformed current breast cancer drugs, leading to further exploration of four additional combinations, three of which also yielded promising results.

This study represents a pioneering closed-loop system where experimental outcomes inform the AI's suggestions, and the AI's outputs guide further experiments conducted by human researchers.

The findings underscore the integration of AI into the scientific discovery process, enabling real-time hypothesis generation and validation based on experimental data.

Researchers emphasize that LLMs like GPT-4 should be viewed as collaborative tools rather than replacements for human scientists, enhancing the drug discovery process.

Professor Ross King highlighted that the collaboration between AI and scientists allows for quicker idea generation, enhancing the overall pace of scientific exploration.

The researchers stress that AI is meant to assist, not replace, human scientists, facilitating faster exploration and testing of new hypotheses.

The research received support from the Alice Wallenberg Foundation and the UK Engineering and Physical Sciences Research Council (EPSRC), highlighting institutional backing for the integration of AI in biomedical research.

The findings were published in the Journal of the Royal Society Interface, showcasing the innovative role of AI in drug discovery.