An Integrated Immuno-Computational Framework for Glioblastoma Therapy: Modeling Synergies in Focused Ultrasound, Nanoparticle-Mediated Immunomodulation, and CAR-Engineered NK/T-Cell Interventions
Authors/Creators
Description
Glioblastoma multiforme remains among the most lethal human malignancies, and its treatment is obstructed by the blood–brain barrier, a profoundly immunosuppressive microenvironment, and pronounced antigenic heterogeneity. This work develops a fully theoretical, computational framework that couples three mechanistically complementary modalities—magnetic-resonance-guided focused ultrasound (FUS) for transient barrier permeabilisation, ligand-conjugated nanoparticles for localised immunomodulation and antigen release, and chimeric-antigen-receptor (CAR) engineered natural-killer and T-cell effectors—within a single quantitative-systems-pharmacology model. The tumour–effector–nanoparticle–antigen dynamics are represented by a four-compartment stochastic partial integro-differential equation (SPIDE) that combines local diffusion, non-local chemotactic recruitment, heavy-tailed antigen dispersal, and multiplicative noise. We integrate the system with a Crank–Nicolson/Milstein scheme on a radial grid, quantify inter-patient variability through a virtual-population layer, apportion output variance with variance-based (Sobol') and screening (Morris) global sensitivity analyses, and infer parameters through an affine-invariant ensemble Bayesian sampler with Watanabe–Akaike information-criterion (WAIC) model comparison. Under the stated modelling assumptions and the illustrative preclinical parameter ranges adopted here, the model predicts that the triadic combination achieves tumour control in 87.0% (95% CI, 84.3–89.7%) of virtual patients, compared with 18.3% (15.3–21.3%) for CAR monotherapy and 0.8% (0.2–1.7%) for the untreated control, a risk difference of 68.7 percentage points (64.5–72.7). Sensitivity analysis identifies CAR-mediated cytotoxic efficacy and FUS-conditioned delivery as the dominant, strongly interacting determinants of response. We emphasise that all reported quantities are outputs of the described model rather than empirical efficacy estimates; they carry no independent evidential weight and are provided solely to characterise the behaviour of the framework and to motivate experimental testing. A complete, seeded, and independently reproducible computational pipeline is included.