In a groundbreaking study, researchers at the National Cancer Institute (NCI), a division of the National Institutes of Health (NIH), have unveiled a pioneering artificial intelligence (AI) tool designed to predict cancer responses to specific drugs. Their findings, published in Nature Cancer, herald a potential shift in cancer treatment personalization through the utilization of single-cell RNA sequencing data.
Traditionally, the process of matching patients with appropriate drugs relied on bulk sequencing techniques, which provided an average representation of all cells within a tumor sample. However, tumors are highly heterogeneous, comprising various cell types and subpopulations, known as clones, each potentially responding differently to drugs. This inherent complexity has often confounded treatment strategies, leading to suboptimal outcomes for many patients. Enter single-cell RNA sequencing—a cutting-edge technology capable of providing unparalleled resolution at the cellular level, offering insights into the intricate dynamics of tumor heterogeneity.
The researchers at NCI embarked on a novel approach, leveraging transfer learning—a machine learning technique—to train their AI model initially on bulk sequencing data and subsequently fine-tuning it using single-cell RNA sequencing data. This strategy aimed to harness the advantages of both approaches: the widespread availability and cost-effectiveness of bulk sequencing data, combined with the superior resolution and granularity offered by single-cell RNA sequencing.
Their investigation encompassed an extensive array of experiments, including the development of AI models for 44 FDA-approved cancer drugs using published cell-line data from large-scale drug screens. Impressively, these AI models demonstrated remarkable accuracy in predicting drug responses at the individual cell level, even for combination therapies—a significant step forward in precision oncology.
Subsequent validation studies on patient data further underscored the utility of the AI tool. Notably, the researchers uncovered that the presence of a single drug-resistant clone within a tumor could render the patient unresponsive to the drug, highlighting the critical importance of targeting specific clones for effective treatment outcomes. Moreover, the AI model successfully anticipated the development of drug resistance in patients with non-small cell lung cancer undergoing targeted therapies—a crucial insight for clinical decision-making.
Despite these promising findings, the researchers acknowledged that the widespread adoption and efficacy of this approach hinge on the increased availability of single-cell RNA sequencing data in clinical settings. To facilitate its translation into practice, they have developed a user-friendly research website and guide for implementing the AI model, dubbed Personalized Single-Cell Expression-based Planning for Treatments In Oncology (PERCEPTION).
Led by Drs. Alejandro Schaffer and Sanju Sinha of NCI, under the supervision of Dr. Eytan Ruppin, the study underscores NIH’s commitment to advancing cancer research and improving patient outcomes. As the nation’s premier medical research agency, NIH, through NCI and its network of institutes and centers, continues to spearhead efforts aimed at unraveling the complexities of cancer and developing innovative strategies to combat this formidable disease.
