Researchers at the University of Pennsylvania’s Perelman School of Medicine and Abramson Cancer Center used an AI, human-in-the-loop framework to nominate and validate a multi-cancer CAR T target. The work, published in Cell, focused on GPNMB as the top antigen candidate after filtering more than 10,000 potential targets from human skin cancer and healthy tissue single-cell RNA-seq datasets. The approach integrates large language models with iterative target nomination and expert review, aiming to reduce hallucinations while preserving scientific selection criteria. In engineered preclinical CAR T models, the GPNMB-directed cells eliminated tumors in melanoma as well as monoblastic leukemia and colorectal adenocarcinoma. The study highlights how AI can compress antigen discovery timelines for solid-tumor CAR T development, a long-standing bottleneck in translating blood-cancer success to more complex tumor microenvironments.