Cancer cells are remarkably good at fighting back. When doctors use drugs to block their growth, the cells often adapt, find a detour, and keep spreading. As a result, finding the right mix of medicines to block a cancer cell’s escape routes is challenging.

Traditionally, scientists test thousands of drug combinations by hand or make an educated guess based on prior research.

Now, researchers at UCLA have built an artificial intelligence platform to ramp up the process.

The tool is called MAESTRO, short for Mechanistic Augmentation and Ensemble Selection for Therapeutic Ranking Optimization. Instead of running endless physical tests, MAESTRO uses large language models to read scientific literature, map out how cancer cells survive, and rank the best drug teams.

“Developing drug combinations currently relies on either brute-force screening or researchers hypothesizing potential candidates based on existing literature,” said Neil Lin, PhD, an associate professor of mechanical and aerospace engineering and bioengineering at the UCLA Samueli School of Engineering and senior author of the study. “It is time-consuming, and it can still be difficult to capture the complex interactions among multiple biological pathways. MAESTRO could substantially accelerate this process. By integrating existing biomedical knowledge, it can identify drug combinations that have a clear biological rationale and are most appropriate for experimental testing.”

Finding Hidden Drug Combinations

Advertisement

Photo: innni/Shutterstock

The UCLA team tested MAESTRO on advanced prostate cancer, a disease that often relies on androgen receptor signaling to grow. A common drug called enzalutamide blocks that signal, but the cancer eventually learns to bypass it.

To find a partner drug for enzalutamide, MAESTRO analyzed 3,592 natural products. It evaluated each candidate over 100 times to make sure the results were consistent, picking 150 top candidates.

When researchers tested those 150 candidates in 3D cancer cell clusters, 23 showed a clear team effort with enzalutamide. One plant compound called berberine stood out by working well across four different resistant cancer models.

“Berberine was not an obvious candidate because it affects several biological pathways, making it difficult to predict how it might work with other drugs,” said James C.H. Lo, first author of the study. “What surprised us was that the combination showed consistent synergy across all of the prostate cancer models, despite their different resistance mechanisms. This is where MAESTRO demonstrated its strength, helping us identify a promising combination that might have been overlooked by examining each pathway separately.”

The researchers plan to test the berberine combination in animal studies next.

“Large language models can be more than conversation or text-generation tools,” Lin said. “With appropriate scientific design and validation, they can support drug discovery and generate hypotheses that can be tested experimentally.”