CatalystMD: AI-Powered Drug Discovery Platform

CatalystMD automates early drug discovery by screening real drug compounds against disease protein targets such as COVID-19 protease, HIV-1 protease, EGFR lung cancer, and KRAS G12C (once considered undruggable for 40 years) using real computational chemistry. Team of 5 AI agents work in sequence, each passing results to the next: 1. Target Analyst Agent: Downloads the real 3D protein structure from the Protein Data Bank, identifies the binding pocket where drugs attach, and analyzes the biological context using Qwen 2.5-7B. 2. Molecular Dynamics Engine: Docks each compound into the binding pocket using AutoDock Vina (physics-based scoring) and runs energy minimization with OpenMM on the AMD MI300X GPU using the AMBER14 force field. 3. Binding Scorer Agent: Ranks all candidates by binding affinity and compares each one to the current FDA-approved drug for that target. Qwen generates a scientific interpretation of the results. 4. Toxicity Screener Agent: Checks toxicity & drug-likeness using Lipinski's Rule of Five and screens for PAINS (Pan-Assay Interference Compounds) using RDKit. 5. Discovery Reporter Agent: Generates a complete scientific brief with structural analysis, rankings, toxicity data, and recommended next steps. Targets tested: COVID-19 Main Protease (6LU7), KRAS G12C Lung Cancer (6OIM), EGFR Kinase Lung Cancer (1M17), HIV-1 Protease (1HIV) Real computation, not mock data: CatalystMD uses real X-ray protein structures from the RCSB Protein Data Bank, real compound SMILES from published literature, AutoDock Vina physics-based docking scores, OpenMM energy minimization, RDKit molecular property calculations, and Qwen 2.5-7B analysis served through vLLM. The AMD MI300X powers both the OpenMM molecular simulation through ROCm OpenCL and AI inference through vLLM. Tech stack: AMD MI300X, AutoDock Vina, Meeko, Open Babel, OpenMM, AMBER14, Qwen 2.5-7B-Instruct, vLLM, LangGraph, RDKit, PDBFixer, FastAPI, Python 3.12, Next.js 16, and 3Dmol.js.