AI-Guided Design of Oral Insulin Analogs Using GNN and ADME Modeling

This project presents an AI-driven computational framework for the rational design of oral insulin analogs capable of overcoming the major challenges of gastrointestinal degradation and poor intestinal absorption. A ten-stage automated pipeline was developed integrating graph neural networks (GNNs), molecular docking, molecular dynamics estimation, gastrointestinal stability modeling, nanoparticle delivery assessment, ADME profiling, and PI3K/AKT pathway safety evaluation. Starting from the native insulin B-chain sequence, 300 mutant analogs were generated through guided stochastic mutagenesis and evaluated using multi-objective optimization. The pipeline identified MUT_271 (L6W;S9A;V12W;G20F) as the top-performing candidate with strong receptor binding affinity, enhanced structural stability, favorable GI resistance, efficient nanoparticle compatibility, and acceptable safety characteristics. Validation metrics demonstrated strong consistency across predictive modules, including significant docking and stability correlations. The study establishes a unified and reproducible AI-based platform for peptide therapeutic engineering and provides a promising computational strategy for future development of orally deliverable insulin therapeutics.