CAR-T Targets

Identification of optimal target antigens remains a key challenge in chimeric antigen receptor (CAR) cell therapy. Bulk expression profiling suffers from the heterogeneity of cancer and normal tissues. In this study, to dissect tissue complexity to the level of individual cells, we constructed a single-cell expression meta-atlas integrating ~1.4 million tumor, patient-derived normal, and reference normal cells of 111 types from 412 tumors of 17 types and 12 normal organs. Deep learning was employed to search this meta-atlas for logical combinations of surface antigens that best discriminate between individual malignant and normal cells. Single-cell epitope profiling validated the AND, OR, and NOT switches in several cancer types. Our resources will facilitate optimized CAR logic design to maximize antitumor efficacy and minimize off-tumor toxicity.

Detailed information of the datasets used for cancer single-cell meta atlas