One of the challenges in understanding how tumors respond to drugs is the inherent complexity of DNA replication — a mechanism targeted by numerous cancer drugs.

“Hundreds of proteins work together in complex arrangements to replicate DNA," Ideker noted. "Mutations in any one part of this system can change how the entire tumor responds to chemotherapy.”

The researchers focused on the standard set of 718 genes commonly used in clinical genetic testing for cancer classification, using mutations within these genes as the initial input for their machine learning model. After training it with publicly accessible drug response data, the model pinpointed 41 molecular assemblies — groups of collaborating proteins — where genetic alterations influence drug efficacy.

“Cancer is a network-based disease driven by many interconnected components, but previous machine learning models for predicting treatment resistance don’t always reflect this,” said Ideker. "Rather than focusing on a single gene or protein, our model evaluates the broader biochemical networks vital for cancer survival."

After training their model, the researchers put it to the test in cervical cancer, in which roughly 35% of tumors persist after treatment. The model was able to accurately identify tumors that were susceptible to therapy, which were associated with improved patient outcomes. The model also effectively pinpointed tumors likely to resist treatment.

Further still, beyond forecasting treatment responses, the model helped shed light on its decision-making process by identifying the protein assemblies driving treatment resistance in cervical cancer. The researchers emphasize that this aspect of the model — the ability to interpret its reasoning — is key to the model’s success and also for building trustworthy AI systems.

"Unraveling an AI model's decision-making process is crucial, sometimes as important as the prediction itself," said Ideker. "Our model's transparency is one of its strengths, first because it builds trust in the model, and second because each of these molecular assemblies we’ve identified becomes a potential new target for chemotherapy. We’re optimistic that our model will have broad applications in not only enhancing current cancer treatment, but also in pioneering new ones." 

Co-authors of the study include: Xiaoyu Zhao, Akshat Singhal, Sungjoon Park, JungHo Kong and Robin E. Bachelder, all at UC San Diego.

This study was supported by the National Institutes of Health (NCI U54 CA274502 and P41 GM103504), Merck KGaA and the Eric and Wendy Schmidt AI in Science Postdoctoral Fellowship, a Schmidt Futures program.

Disclosures: Trey Ideker is cofounder, consultant and has equity in Serinus Biosciences and Data4Cure. He is also a consultant and has equity in Ideaya Biosciences.