Creating a flexible, scalable, and reproducible automated pipeline that can handle the entire process, from a protein’s sequence to a list of prioritized leads, can help reduce the notoriously lengthy timelines for new medicines to reach the market. This pipeline involves integrating various complex, state-of-the-art computational tools for molecular dynamics, modeling, and docking, while also ensuring the pipeline can function even when a protein’s crystal structure isn’t available.
Why is this project particularly interesting for BioExcel?
This project is interesting because it merges several high-impact computer-aided drug discovery techniques into biomolecular workflows relevant to the pharmaceutical industry. It offers an end-to-end solution for drug design, allowing researchers to move from a protein sequence to a set of prioritized ligand binding poses with minimal manual intervention. The pipeline is built on the BioExcel Building Blocks (BioBB), a collection of modular and interoperable wrappers around state-of-the-art biomolecular simulation tools. Guided by FAIR4RS principles and usability, its iterative refinement through active industrial projects with partners like Nostrum Biodiscovery proves its real-world applicability, robustness, and readiness for actual drug discovery projects compared to standard tutorials.
What are we doing in BioExcel?
We are developing and showcasing an automated drug design pipeline using the BioBB ecosystem, which is divided into Molecular Dynamics (MD) simulation workflows and Virtual Screening (VS) workflows.
- The process begins with either a protein sequence or an existing structure. If an experimental structure isn’t available, we use AlphaFold or another structure prediction model starting from a sequence (e.g., from UniProt) to generate wild-type or mutant structures de novo. If available, we download it via PDB code or read a local file.
- Once a target structure is available, the Protein Preparation workflow fixes PDB defects and assigns protonation states with PROPKA3 using tools like biobb_io, biobb_model, biobb_structure_utils, biobb_pdb_tools, biobb_amber, and biobb_chemistry.
- We then run molecular dynamics (MD) simulations using GROMACS (including energy minimization, NVT/NPT equilibration, and production runs with or without PLUMED), followed by comprehensive trajectory post-processing and basic analysis. Replicas and systems can be launched via the Horus workflow manager. Advanced analysis and clustering via Deep Cartograph identify slow modes, compute free energy surfaces, and extract representative conformations.
- These representative conformations serve as a basis for virtual screening and docking. Using fpocket to find pockets and Autodock Vina (or HADDOCK3 for more refined docking), we screen an input SDF ligand library using configurable CPU and exhaustiveness settings.
- The final output is a ranked list of ligands and PDB files containing top-ranked binding poses that can be used for further analysis or experimental validation.
You can obtain further information on both the MD and VS workflows via the corresponding repositories that include documentation and tutorials runnable either locally or in Google Colab.
