Drug discovery has traditionally been constrained by time-intensive and cost-prohibitive experimental screening. The proposed study introduces an Explainable Generative Multi-Objective Framework (EGMOF) for end-to-end drug discovery. The framework combines four major modules: (1) Data Preprocessing and Representation using MolFormer for molecular embeddings and ProtT5 for protein sequence encoding. (2) Generative Design employing a Graph Transformer-VAE hybrid model for de novo molecule generation guided by learned chemical constraints. (3) Multi-Objective Optimization utilizing Reinforcement Learning with Pareto Front Approximation (RL-PFA) to balance ADMET, binding affinity, and drug-likeness scores. (4) Explainable Prediction and Validation through SHAP-XAI and AlphaFold-Docking Ensemble (AF-DE) for interpretable interaction mapping and binding energy estimation. The combination MolFormer, Graph Transformer-VAE, RL-PFA, and SHAP-XAI with AF-DE, the proposed framework achieves superior performance: binding affinity ΔG: –11.4 kcal/mol, Ki: 110 nM, IC50: 165 nM, QED: 0.81, drug-likeness: 0.78, docking success rate: 97%, significantly outperforming existing generative AI methods in multi-objective drug discovery.
