Researchers from the Perelman School of Medicine and the Abramson Cancer Center at the University of Pennsylvania have developed a human-in-the-loop artificial intelligence (AI) framework to streamline the identification of viable target antigens for CAR T cell therapy. Finding suitable antigens for cancers beyond blood malignancies has traditionally been a significant challenge, requiring manual data curation methods that can take several months to several years. Published in the journal Cell, the team’s customizable workflow successfully compressed this discovery phase down to less than a few weeks.
The computational strategy combines large language models (LLMs) with single-cell RNA sequencing data to handle massive amounts of genomic information. This system centers the existing expertise of scientists, allowing them to effectively validate antigen candidates nominated by the AI. As a proof-of-concept, the platform identified glycoprotein non-metastatic melanoma protein B (GPNMB) as its top candidate. In preclinical laboratory testing, engineered GPNMB-targeted CAR T cells demonstrated robust tumor-killing efficacy in mouse models, achieving remission without relapse across multiple distinct cancer types, including melanoma, leukemia, and colorectal cancer.