AI Tool created by NIH Researchers Promises Enhanced Precision in Matching Cancer Drugs to Patients

April 21, 2024

In a recent proof-of-concept study, NIH researchers have unveiled an AI tool designed to predict cancer drug responses by analyzing data from individual tumor cells. Published in Nature Cancer on April 18, 2024, by researchers from the National Cancer Institute (NCI), this study suggests that leveraging single-cell RNA sequencing data could revolutionize the process of matching cancer patients with effective drugs.

Traditional methods for matching patients with drugs rely on bulk sequencing of tumor DNA and RNA, which provides an averaged overview of all cells within a tumor sample. However, tumors consist of diverse cell populations, including various subpopulations known as clones, which may exhibit distinct responses to drugs. These differences in response could explain why certain patients fail to respond to specific drugs or develop resistance.

In contrast, single-cell RNA sequencing offers significantly higher resolution data, enabling the identification and targeting of individual clones. Despite its potential, single-cell gene expression data are currently more expensive to generate and not widely accessible in clinical settings.

To address this, the researchers explored the feasibility of using transfer learning, a machine learning technique, to train an AI model initially using bulk RNA sequencing data and then refining it with single-cell RNA sequencing data. This approach, applied to published cell-line data from large-scale drug screens, allowed the researchers to develop AI models for 44 FDA-approved cancer drugs. These models accurately predicted how individual cells would respond to both single drugs and drug combinations, indicating the potential of this method to enhance personalized cancer treatment strategies.