Artificial intelligence for overcoming resistance and safety challenges in cancer stem cell-targeted cellular immunotherapy

Abstract Cancer stem cells (CSCs) represent a dynamic tumor cell state associated with self-renewal, phenotypic plasticity, disease recurrence, and therapeutic resistance. CSC-targeted cellular immunotherapy, including chimeric antigen receptor (CAR)-based approaches, is therefore constrained by heterogeneous antigen expression, phenotypic switching, tumor microenvironmental suppression, and the risk of on-target, off-tumor toxicity. These challenges are particularly pronounced in solid malignancies, where limited immune-cell trafficking, hypoxia, metabolic competition, and stromal barriers complicate treatment compared with hematological cancers. Artificial intelligence (AI) gives a framework for integrating high-dimensional molecular, cellular, spatial, and clinical data to address these limitations. Single-cell and multi-omics machine learning can identify stemness-associated states and resistance programs that may be missed by conventional single-marker approaches, while predictive models can prioritize tumor-selective antigens and anticipate treatment resistance or toxicity. Computational protein-design approaches further enable optimization of CAR antigen-binding domains, with emerging experimental studies demonstrating improved receptor performance using AI-designed binders. AI may also support multi-antigen and logic-gated CAR selection and longitudinal monitoring of CSC-state transitions during treatment. However, most current applications remain computational or preclinical, and clinical utility has not yet been established prospectively. This review examines AI strategies for overcoming CSC-associated resistance and improving therapeutic specificity and safety, while distinguishing computational, preclinical, retrospective clinical and prospective clinical evidence. Integrating AI with molecular engineering, experimental validation, and prospective clinical evaluation may enable more adaptive, selective and safer CSC-targeted cellular immunotherapies. Particular emphasis is placed on disease-specific modeling, tumor-versus-normal discrimination, resistance-network integration, external validation and reproducibility as prerequisites for responsible clinical translation across diverse patient populations.