Automated Drug Design Pipeline Integrating Deep Learning-Based Docking and Molecular Dynamics

We present an end-to-end computational pipeline for structure-based drug design that integrates deep learning models with physics-based refinement. The workflow begins with generative ligand design using diffusion-based models, followed by a two-stage docking process combining rigid and flexible side-chain sampling. An AI-driven interface manages preprocessing, ligand selection, and parameter tuning through natural-language commands, allowing dynamic adjustment of pipeline operations. Final refinement is achieved through molecular dynamics simulations and energy minimization, producing physically realistic binding conformations. Key innovations include automated protein–ligand handling with minimal user intervention, improved conformational sampling during docking, and physics-grounded refinement for pose accuracy. Benchmarking across diverse protein targets shows enhanced computational efficiency and prediction accuracy compared to traditional approaches. Supported in part by QIS@Perlmutter, this work also explores quantum-inspired physical strategies for sampling conformational energy landscapes, strengthening its relevance to computational biophysics.