Development of Molecular and Quantitative Spatiotemporal Modeling of Physics Informed Neural Network (PINN) with in Vitro-Driven Validation Modeling for Blood-Tumor Barrier Transport, Resistance Dynamics, and Therapeutic Penetration Mechanisms in Patient-Specific Glioblastoma Surgery

Author(s): Shivi Kumar, Deirdre Richardson, Osama Elzafarany, Teryn Mitchell, Katheryn Dampos

Glioblastoma Multiforme (GBM) remains the most aggressive primary brain malignancy, characterized by profound therapeutic resistance and near-universal recurrence despite maximal surgical resection and chemoradiation. The central obstacle to effective treatment lies in the structural and molecular complexity of the blood-brain barrier (BBB), which severely limits drug delivery, and the tumor’s adaptive evolution that promotes immune evasion and metabolic resilience. Current therapeutic strategies treat GBM as a homogeneous mass, failing to account for the spatial and temporal heterogeneity that defines its resistance to chemotherapy. This study introduces an integrative, biologydriven spatiotemporal modeling framework designed to map and predict drug transport resistance across the BBB and within tumor subregions, enabling patient-specific optimization of neurosurgical and pharmacologic intervention. The model fuses transcriptomic, radiologic, and biophysical data to replicate the dynamic interplay between endothelial permeability, efflux transporter expression, cytokine signaling, and immune infiltration. Using high-resolution datasets from TCGA-GBM, CPTAC, and YAIB, over 14,000 data points encompassing microvascular density, astrocytic activation, and efflux kinetics were embedded into a physics-informed system governed by diffusion–reaction partial differential equations. This approach allows simulation of molecular flux across heterogeneous tumor environments, reproducing the observed gradients of drug penetration failure at invasive margins and hypoxic zones. Through quantitative coupling of radiogenic parameters and molecular biomarkers—including MGMT methylation, IDH1/2 mutation status, HIF1α induction, and VEGFdriven neoangiogenesis—the model identifies specific resistance collapse points: regions where therapeutic efficacy diminishes due to cumulative mechanical and metabolic constraints. Validation was achieved through cross-modality comparison between model-predicted resistance maps and patient MRI follow-ups, yielding a mean spatial concordance of 0.87 and predictive accuracy exceeding 92 percent. These results collectively reveal that therapeutic failure in GBM arises not merely from pharmacologic inadequacy, but from spatiotemporal synchronization between cellular plasticity and microvascular dysfunction. By translating this computational insight into a predictive biological framework, this research establishes the foundation for individualized surgical targeting and optimized drug infusion strategies in glioblastoma. Future experimental phases will expand this work through microfluidic BBB-on-chip systems and radiogenomic datasets to refine predictive capacity.