How drewnutt/open_combind works
Main Components
• Input/Curation Component – This includes data curation and preparation of raw inputs. The pipeline expects structural data (protein and ligand files) in specific directories (e.g. structures/raw, structures/proteins, structures/ligands). The “structprep” and “ligprep” commands in the README demonstrate how raw data is first curated.
• Docking Engine – Implemented in “open_combind/dock”, this module wraps the docking process using GNINA as the underlying docking tool. It handles tasks like generating docking template files and executing GNINA commands.
• Scoring & Pose Prediction – The “open_combind/score” module, combined with “open_combind/pose-prediction” and “extract-top-poses” commands, uses the computed features to predict the best binding poses and rank them.
• Command Line Interface (CLI) – The “open_combind/cli” folder indicates available CLI tools to drive commands like “structprep”, “ligprep”, “dock-ligands”, “featurize”, “pose-prediction”, and “extract-top-poses”.
• Supporting Modules & Data – Additional folders such as “open_combind/chembl”, “open_combind/data”, “open_combind/pymol” (for visualization) and “open_combind/stats_data” (for benchmarking statistics) support core functions and provide auxiliary data sets.
• Testing & Documentation – “open_combind/tests” ensures functionality and “docs” provides user and developer insights into the API and usage instructions.
Depicting Relationships and Interactions
• That these inputs are split into two parallel processes: one responsible for preparing the protein structures (via “structprep”) and one for processing ligand information (via “ligprep”).
• The docking results then move into the Featurization Module (open_combind/features) where pose features are computed.
• Next, these features are used by the Pose Prediction/Scoring Module (open_combind/score) to determine and extract the optimal poses.
• Finally, include paths for benchmarking and validation using modules in “open_combind/stats_data” and tests in “open_combind/tests”.
Important Architectural Patterns/Design Principles
• Modular design: Each module (docking, featurization, scoring) is self-contained and can be invoked independently via CLI commands.
• Pipeline workflow: Emphasize that the overall process is sequential with specific hand-offs between components.
• Extensibility: The use of “open_combind/cli” and segregated module directories demonstrates an architecture that anticipates extensions (for example, swapping out docking algorithms or integrating new featurization techniques).
• Integration of external tools: Clearly denote GNINA, OpenMM, PDBFixer, PyMol, and ProDy as key external dependencies that interact with internal modules.
Listing Relevant Technologies and Tools
• Languages/Frameworks: Python (as indicated by the .py files and setup.py/config files)
• External Dependencies: GNINA, OpenMM, PDBFixer, PyMol (mentioned explicitly in the README), ProDy.
• Build/Deployment Tools: Provided by “pyproject.toml”, “setup.cfg”, and scripts in “devtools/” for environment creation and setup.
• Continuous Integration/Documentation: The presence of GitHub workflows, Codecov, and ReadTheDocs integration shows emphasis on testing and documentation.