How drorlab/combind works
Main Components
• Data Curation & Preparation: Show how raw structural data and ligand files enter the pipeline. Note that the README details requirements for a “structures/raw” directory (even though it isn’t in the provided tree, it is implicitly expected) and how split files (protein and ligand) are managed.
• Structural Preprocessing: Include a module for structure preparation (e.g., the “chembl” directory, which likely handles data curation for chemical structures, and “dock” for structural alignment and grid generation).
• Ligand Preparation (LigPrep): Clearly represent the ligand processing stage (this is invoked via the “combind ligprep” command).
• Docking: Represent the docking component as a dedicated module (from the “dock” directory) where the prepared ligands are docked into a protein grid.
• Scoring & Pose Prediction: Include a module for “score” where the predicted poses are compared, scored, and benchmarked.
• Visualization: Represent the “pymol” component that handles visualization and interaction analysis (often used to view complexes and poses).
• Benchmarking & Statistics: Add a separate component to indicate benchmarking and reference statistics (using data in the “stats_data” directory).
Relationships and Data Flow
• Move to Docking: From the preprocessing nodes, show data flows into the Docking stage (the “dock” modules) where a docking grid and docked poses are generated.
• Proceed to Scoring & Pose Prediction: From featurization, indicate that data flows into the scoring modules (score directory) which compute binding affinities and predict the best poses.
• Optional Virtual Screening (ComBindVS): Include a branch for the ComBindVS workflow where a similar but adjusted pipeline (with options like --multiplex and --screen) is applied for virtual screening.
• Finally, Visualization & Post-processing: Finally, show an optional path that flows into visualization (using pymol modules) and external Schrodinger utilities (such as glide_sort) for final analysis.
Architectural Patterns and Design Principles
• Pipeline-Based Modular Architecture: Emphasize that each step (curation, docking, featurization, scoring) is a separate module that can be executed individually or as part of an end-to-end pipeline.
• Command-Line Orchestration: Note that the tool is controlled via command-line instructions (“combind <subcommand> …”) which ties the modules together.
• Integration with External Tools: Highlight that external libraries and tools (Schrodinger’s Glide, Schrodinger Python API, numpy, scipy, pandas, lmdb, etc.) are essential parts of the system.
• Test and Benchmark Components: Recognize that testing (as seen in score/tests) and benchmarking (stats_data) are integral for ensuring reliability and performance.