DeepPurpose-MD: An End-to-End Colab-Based Ligand Discovery Pipeline Integrating Docking, Molecular Dynamics, and Deep Learning

DeepPurpose‑MD v2.5 is a unified, fully executable pipeline for molecular discovery that integrates deep learning–based target screening, molecular docking, and GPU‑accelerated molecular dynamics (MD)—all inside an interactive Google Colab notebook. The workflow runs end‑to‑end (dataset → DTI screening → docking → MD → post‑MD analysis) with reproducible defaults and zero local setup. This work describes an in-silico computational pipeline and does not provide medical advice, diagnosis, or treatment recommendations. What’s new in v2.5 (Oct 2025) — incremental reliability update Reproducibility: added a --seed flag to the MD script (OpenMM) and a seed flag to docking (AutoDock Vina) so users can fix or sweep random seeds across runs. Better docking control: added an optional manual box‑center override (x/y/z) while keeping automatic centroid strategies (protein atoms, RNA residue range, HETATM centroid). Input sanitation: pre‑processing now removes duplicated atoms in receptor inputs to avoid downstream force‑field failures. Apo vs. Holo parity: standardized defaults (e.g., 7.0‑nm water box, 0.15 M ions) and simplified apo/holo comparison switches for fair, side‑by‑side analyses. Robust logging & errors: clearer console messages and early checks so users see actionable hints instead of opaque stack traces. v2.0 capabilities retained Protein and RNA targets: automated preprocessing and force‑field selection (including RNA‑specific support) with optional RNA analyses. Docking → MD continuity: best pose extraction, Kabsch alignment, and conversion to SDF for consistent ligand coordinates between docking and MD. Post‑MD analysis UI: code‑free toggles in Colab for common analyses (RMSD/RMSF overlays, PCA/FEL, hydration shell, H‑bonds, contact maps) with publication‑ready figures and CSV/TSV exports. DeepPurpose (enhanced): a customized fork with ProtTrans embeddings + PCA compression (e.g., Pearson ≈ 0.80 on BindingDB‑COVID), used for rapid DTI pre‑screening before docking/MD. Who is this for?Researchers and practitioners in cheminformatics, structural biology, and AI‑driven small molecule design who want a reproducible, cloud‑friendly pipeline that bridges ML screening with physics‑based refinement—without maintaining a local HPC stack. Highlights One‑click Colab notebooks, modular scripts, and clear CLI flags. Fully automated apo/holo side‑by‑side MD and analysis. Reproducible seeding across docking and MD. Exportable figures (PNG/PDF) and tables (CSV/TSV) for manuscripts or ELNs. Getting started Open the Colab notebook, follow the top‑cell checklist, and run the pipeline end‑to‑end. Swap targets/ligands via the provided cells and CLI flags; seed, box, and docking center are configurable. License & citationPlease cite this Zenodo record when using the pipeline in academic or industrial work. See repository LICENSE for terms.