A Hierarchical Bayesian Multiscale Framework for an Immuno-Regenerative Nanoplatform Targeting Beta-Cell Mass Restoration in Paediatric Type 1 Diabetes: Design Criteria, Identifiability and Falsifiable Predictions

Type 1 diabetes (T1D) in children arises from autoimmune destruction of insulin-producing β-cells. Disease-modifying immunotherapy slows β-cell loss without restoring mass, whereas stem-cell-derived islet replacement restores endogenous insulin secretion but currently requires chronic immunosuppression. A therapy that simultaneously re-establishes antigen-specific tolerance and regenerates β-cell mass would close this gap. We formulate and analyse, entirely in silico, a dual-target nanoplatform combining a thermoresponsive chitosan–poly(N-isopropylacrylamide) depot releasing interleukin-10 and a proinsulin peptide with ionisable lipid nanoparticles bearing the epidermal-growth-factor-receptor-binding peptide GE11 and encoding PDX1, MAFA and NEUROG3 for in vivo reprogramming of pancreatic acinar cells. A thirteen-state, dimensionally consistent multiscale model couples depot erosion, nanoparticle pharmacokinetics, transcript kinetics, dendritic-cell mediated regulatory T-cell induction with infectious-tolerance feedback, latent viral reactivation, autoantigen balance and β-cell mass, and is linked to clinical endpoints through a glucose–insulin submodel and the ADAG glycated haemoglobin relationship. The integrator is verified to fourth order (observed orders 3.996, 4.001, 4.091) against an analytic pharmacokinetic solution and by grid refinement. Three results follow. First, the tolerance subsystem is bistable: the regulatory T-cell nullcline has three roots (4.68, 6.64, 54.71 cells mm⁻³), so a transient tolerogenic stimulus produces a permanent switch only if it crosses the separatrix; deleting the infectious-tolerance feedback reduces the one-year β-cell mass from 0.580 to 0.249. Second, prior-predictive propagation over nine literature-informed priors (n = 16384) gives a probability of 98.8% (95% CI 98.6–99.0) for stimulated C-peptide above 0.5 ng mL⁻¹ at one year, but only 14.6% (14.1–15.2) for HbA1c below 6.5% at five years, because durability is limited not by immunity but by the stability of the reprogrammed phenotype: a durable functional cure requires the loss rate of reprogrammed β-like cells to satisfy k_d ≤ 2.1×10⁻⁴ d⁻¹, a half-life of at least 8.9 years. Third, time-resolved Sobol' analysis shows that output variance is initially controlled by conversion and delivery (S_T^Vmax = 0.442, S_T^ε = 0.400 at week 4) and that the qualitative change by week 52 is the emergence of phenotype durability, S_T^kd rising from 0.002 to 0.216 within an ordering led by targeting (0.374) and conversion (0.273), while immune parameters carry almost no variance despite being structurally essential. The framework yields explicit, quantitative and falsifiable design criteria rather than efficacy claims, and every reported number is reproduced by the embedded code.