Researchers used an AI human-in-the-loop framework to nominate CAR T targets and reported preclinical activity for a GPNMB-directed construct across multiple cancer models. In a Cell paper, a University of Pennsylvania team led a strategy that integrates large language models with single-cell RNA sequencing datasets to filter and refine antigen lists for CAR T design criteria. The team performed large-scale target nomination simulations to reduce model noise and then selected glycoprotein non-metastatic melanoma protein B (GPNMB) as the top candidate. Engineered GPNMB CAR T cells showed tumor elimination not only in melanoma models but also in monoblastic leukemia and colorectal adenocarcinoma preclinical settings. The work targets a key bottleneck in solid-tumor CAR T development—finding safe, selective antigens—and provides a computational-to-experimental pipeline that could be reused across additional target discovery programs.