Artificial intelligence-driven precision immunogenetics for cancer stem cell-targeted CAR immunotherapy

Cancer stem cells (CSCs) are a major contributor to therapeutic resistance, tumour recurrence, and disease progression because of their self-renewal capacity, phenotypic plasticity, immune evasion, and ability to persist after treatment. Chimeric antigen receptor (CAR)-based immunotherapy has transformed the treatment of several hematological malignancies, but effective targeting of CSC-driven disease, especially solid tumour, remains challenging. Antigen heterogeneity and loss, CSC plasticity, CAR-cell dysfunction, metabolic constraints, and an immunosuppressive tumour microenvironment (TME) can limit long-term therapeutic responses. CSC-associated antigens such as CD133, CD44, EpCAM, HER2, CSPG4, ALDH, and EGFRvIII have been investigated as CAR targets, although their heterogeneous and context-dependent expression raises important concerns regarding CSC specificity and on-target/off-tumour toxicity. Artificial intelligence (AI) and computational approaches provide opportunities to integrate genomic, transcriptomic, proteomic, spatial, and clinical data for CSC-associated antigen discovery, biomarker identification, CAR design, patient stratification, and toxicity prediction. However, the evidence supporting these applications varies considerably, ranging from computational and preclinical studies to retrospective clinical modelling, while prospective clinical validation of AI-guided CSC-targeted CAR therapy remains limited. Similarly, next-generation platforms, including CAR-natural killer (CAR-NK) cells, CAR-macrophages (CAR-M), gamma-delta CAR-T cells, and CAR-invariant natural killer T (CAR-iNKT) cells, remain at different stages of translational development. This review critically examines CSC biology, mechanisms of resistance to CAR-based therapy, emerging CAR platforms, and the potential role of AI in precision immunogenetics. Particular emphasis is placed on differentiating CSC-intrinsic, CAR-cell-intrinsic, and TME-mediated resistance and on defining the evidence maturity of AI applications. Finally, current challenges in data availability, tumour heterogeneity, model bias, external validation, interpretability, safety, and prospective clinical translation are discussed, together with priorities for developing clinically validated AI-enabled CSC-targeted CAR therapies.