Multimodal AI modeling of T cell therapies to predict patient response and nominate advanced cell design strategies

ABSTRACT This project will develop novel multimodal artificial intelligence (AI) platforms to predict clinical outcomes for patients treated with chimeric antigen receptor (CAR) T cell therapies and nominate molecular engineering strategies to enhance therapeutic potency and safety. We will build two complementary AI systems: (i) tcellGPT, which integrates single-cell multi-omic, perturbation, and clinical datasets to predict patient response and toxicity based on the molecular profile of infused CAR T cells, and (ii) tnicheAI, which maps the spatial organization and functional state of the tumor microenvironment in multi-modal histopathology images and integrates clinical data to identify suppressive niches that must be overcome by next-generation CAR T cell designs. These models will be fused into tcellAI to holistically predict patient outcomes by combining infusion product profiles, tumor characteristics, and clinical covariates. Key outcomes will include predictive tools to personalize CAR T cell therapies, accelerate preclinical development of novel cell engineering strategies, and elucidate biological mechanisms underlying treatment efficacy and toxicity. Clinical team and bioethicists will be engaged throughout to proactively address fairness, privacy, transparency, and accountability considerations in the AI development lifecycle. The work will be conducted by a diverse team spanning immuno-oncology, machine learning, clinical care, and bioethics to thoughtfully curate datasets, tailor architectures, and validate predictive insights in alignment with stakeholder needs. Ultimately, by enabling more effective, personalized, and equitable cell therapies, this ethically-grounded multimodal AI approach will bring curative treatments to more patients.