GAN-enhanced digital twins for biomarker-guided cancer immunotherapy and adoptive cell therapy: a mini review.

Cancer immunotherapy and adoptive cell therapy have transformed modern oncology, yet durable benefits remain limited by tumor heterogeneity, immune escape mechanisms, biomarker variability, and the dynamic remodeling of the tumor immune microenvironment. Current biomarker strategies, including PD-L1 (programmed death-ligand 1) expression, microsatellite instability (MSI), tumor mutational burden (TMB), immune gene signatures, liquid biopsy measurements, and target antigen profiles, provide clinically useful information but are often static and insufficient to capture longitudinal tumor-immune-therapy interactions. Digital twin technology offers a computational paradigm for integrating multi-omics, immune profiling, clinical variables, imaging, and treatment-response data into patient-linked virtual models. Generative artificial intelligence, including generative adversarial networks (GANs), may enrich sparse biomarker-defined state spaces and support the simulation of underrepresented treatment response or toxicity trajectories. This mini review critically compares mechanistic quantitative systems pharmacology, spatial and agent-based modeling, data-driven prediction, and emerging digital twin approaches and proposes a platform-agnostic architecture that can be implemented in MATLAB/Simulink or other computational environments. The adoptive cell therapy scope focuses on chimeric antigen receptor T-cell (CAR-T), chimeric antigen receptor natural killer cell (CAR-NK), and T-cell receptor-engineered T-cell (TCR-T) therapies, with related principles applicable to tumor-infiltrating lymphocyte (TIL) therapy. We describe conditional and longitudinal GAN strategies, dynamic tumor-immune state modeling, biomarker integration, validation metrics, governance, and prospective implementation pathways. The proposed framework is conceptual and has not yet been clinically validated; synthetic data are treated as methodological support rather than clinical evidence. Its intended near-term uses are hypothesis generation, virtual-cohort enrichment, trial design, and uncertainty-aware decision support.