Introduction: Pancreatic β-cells dynamically sense extracellular glucose fluctuations and regulate insulin secretion through a closed-loop feedback mechanism. Replicating this autonomous biological behavior within synthetic biomaterials remains a major challenge. This study aimed to develop an artificial intelligence (AI)-optimized, pancreatic islet-inspired self-healing polymer network capable of glucose recognition, adaptive structural reconfiguration, and autonomous modulation of antidiabetic drug release.
Methods: A dynamic polymer network was engineered using phenylboronic acid-functionalized chitosan and oxidized dextran, exploiting reversible boronate ester and Schiff-base interactions to reproduce adaptive biological responsiveness. Metformin-loaded nanodomains and the glucagon-like peptide-1 receptor agonist exendin-4 were incorporated for complementary therapeutic action. A hybrid machine-learning framework integrating Gaussian process regression and Bayesian optimization was employed to optimize polymer composition, crosslinking density, glucose sensitivity, self-healing efficiency, and dual-drug release. Material performance was evaluated under normoglycemic (5 mM), moderate hyperglycemic (15 mM), and severe hyperglycemic (25 mM) conditions.
Results: The optimized network exhibited a self-healing efficiency of 94.6 ± 1.8% within 30 min and recovered 91.3 ± 2.4% of its initial storage modulus following three disruption–recovery cycles. Glucose-dependent network adaptation increased the equilibrium swelling ratio from 486 ± 31% at 5 mM to 823 ± 47% at 25 mM glucose. Cumulative therapeutic release at 72 h increased from 38.7 ± 2.6% under normoglycemic conditions to 81.9 ± 3.4% under severe hyperglycemia, demonstrating a 2.1-fold glucose-dependent response. The AI-guided strategy reduced experimental screening requirements by 73% compared with full-factorial optimization.
Conclusions: This pancreatic islet-inspired polymer network integrates biological feedback principles, dynamic covalent chemistry, autonomous glucose responsiveness, and AI-guided materials optimization within a single adaptive platform. The system provides a promising biomimetic strategy for next-generation closed-loop therapeutic delivery and demonstrates how biological sensing mechanisms can guide the engineering of intelligent, self-regulating materials.