AI Tool Predicts Cancer Therapy Responses Using Tumor Cell Data

With over 200 types of cancer, each with its unique characteristics, the endeavor to develop precise oncology treatments remains formidable. The primary focus has been on crafting genetic sequencing assays or analyses to pinpoint mutations in cancer driver genes and then attempting to align treatments that could combat these mutations.

However, many cancer patients, if not the majority, derive limited benefit from these initial targeted therapies. In a recent study published in the journal Nature Cancer, spearheaded by Sanju Sinha, Ph.D., an assistant professor in the Cancer Molecular Therapeutics Program at Sanford Burnham Prebys, along with senior authors Eytan Ruppin, M.D., Ph.D., and Alejandro Schaffer, Ph.D., at the National Cancer Institute (NCI), part of the National Institutes of Health (NIH), and their colleagues, a pioneering computational pipeline was unveiled to predict patient response to cancer drugs at the level of individual cells.

Termed as PERsonalized Single-Cell Expression-Based Planning for Treatments in Oncology (PERCEPTION), this novel artificial intelligence-driven approach delves deeper into the realm of transcriptomics—the study of transcription factors, which are messenger RNA molecules expressed by genes that carry and convert DNA information into action.

Explore the Upcoming Oncology Conferences and Online Courses to stay updated with the latest advancements in cancer diagnosis and treatment

“A tumor is a complex and evolving entity. Employing single-cell resolution enables us to address both of these challenges,” remarks Sinha. “PERCEPTION harnesses the wealth of information contained within single-cell omics to comprehend the clonal architecture of the tumor and monitor the emergence of resistance.” (In biology, omics refers to the totality of constituents within a cell.)

Sinha elaborates, “The ability to track the emergence of resistance is particularly intriguing for me. It holds the promise of enabling us to adapt to the evolution of cancer cells and potentially adjust our treatment approach.”

Sinha and his colleagues leveraged transfer learning—a branch of artificial intelligence—to develop PERCEPTION.

“Our most significant hurdle was the scarcity of single-cell data from clinical settings. An AI model necessitates copious amounts of data to grasp a disease, similar to how ChatGPT relies on vast quantities of text data extracted from the internet,” explains Sinha.

PERCEPTION utilized published bulk-gene expression data from tumors to pre-train its models. Subsequently, the models were fine-tuned using limited single-cell data from cell lines and patients.

PERCEPTION was successfully validated by predicting responses to monotherapy and combination treatments in three independent clinical trials recently published for multiple myeloma, breast cancer, and lung cancer. In each instance, PERCEPTION effectively stratified patients into responder and non-responder categories. Notably, in lung cancer, it even elucidated the emergence of drug resistance as the disease progressed, marking a significant discovery with profound implications.

Sinha underscores that although PERCEPTION is not yet prepared for clinical deployment, the approach demonstrates that single-cell information can be harnessed to guide treatment decisions. He aspires to foster the adoption of this technology in clinical settings to generate more data, which can subsequently be utilized to refine and advance the technology for clinical application.

“The accuracy of the predictions escalates with the caliber and volume of the foundational data,” notes Sinha. “Our objective is to devise a clinical tool capable of forecasting the treatment response of individual cancer patients in a systematic, data-centric manner. We anticipate that these findings will catalyze the accumulation of more data and the initiation of additional studies, sooner rather than later.”

For more information: PERCEPTION: Predicting patient treatment response and resistance via single-cell transcriptomics of their tumors, Nature Cancer (2024), https://dx.doi.org/10.1038/s43018-024-00756-7