To develop an AI-assisted framework for identifying new targets for CAR T-cell therapy, specifically focusing on GPNMB as a potential multi-cancer target.
Approach:
Methodology: Utilized a human-in-the-loop strategy combining public single-cell RNA sequencing datasets, large language models, expert review, and experimental validation.
Target Identification: Filtered over 10,000 potential targets based on tumor expression, tissue specificity, and clinical feasibility, prioritizing candidates through simulations to minimize model output instability.
Experimental Validation: Confirmed GPNMB surface expression across various tumor types and engineered GPNMB-directed CAR T cells, demonstrating antitumor activity in mouse models.
Key Findings:
GPNMB was identified as a promising target for CAR T-cell therapy.
The framework effectively integrates large datasets and AI to streamline target discovery.
GPNMB CAR T cells exhibited antitumor activity in melanoma, monoblastic leukemia, and colorectal adenocarcinoma in mouse models.
Interpretation:
Limitations:
Further assessment of specificity, safety, manufacturing feasibility, and clinical translation of GPNMB CAR T cells is required.
Conclusion:
The modular and disease-agnostic nature of the framework suggests it could be applicable to other cancers or diseases as datasets and models evolve.
