chem-docking-void

Goal

To perform molecular docking of a small-molecule ligand into a porous material structure (CIF format) using the VOID library. This skill aims to automatically generate a robust sampling of guest conformers using RDKit, optimize them, and then distribute them throughout the host framework using Voronoi-based cluster sampling and physics-informed collision filtering.

This will output:

• Ranked docked complexes saved individually as standard CIF files.
• A metadata summary (docking_results.json) capturing the generation parameters, associated RDKit conformer energies, and matched pose IDs.

Instructions

1. Identify Inputs

You will need:

• The SMILES string of your guest molecule.
• The CIF file path to your porous material (e.g. Zeolites, MOFs).

2. Basic Docking Run

A standard run accepts the chemical inputs and saves outputs to a designated folder.

# Env: atomistic-agent
python .agents/skills/chem-docking-void/scripts/run_docking.py \
  --smiles "CC12C3C4C5C6C1C7C2C3C4C5C67" \
  --host_cif /path/to/host/material.cif \
  --output_dir output/docked_poses \
  --num_conformers 5

(The SMILES here represents Adamantane or similar structures for testing.)

3. Tuning Hyperparameters

The clustering map and acceptance rates are highly sensitive to VOID's search parameters. Use the advanced arguments for dense loading or strict spatial tolerances:

python .agents/skills/chem-docking-void/scripts/run_docking.py \
  --smiles "CC(=O)Oc1ccccc1C(=O)O" \
  --host_cif /path/to/host/MOF.cif \
  --output_dir output/docked_poses \
  --num_conformers 10 \
  --threshold 1.8 \
  --attempts 2000 \
  --structs_per_loading 5 \
  --num_clusters 150 \
  --max_loading 1 \
  --max_subdock 200 \
  --remove_species "H2O" "Na"

Meaning of Key Hyperparameters:

• --num_conformers: (RDKit) How many of the lowest-energy 3D geometries to test.
• --threshold: The acceptable minimum distance (Å) between the host atoms and guest atoms. A lower value allows tighter squeezes but risks atomic clashes.
• --attempts: How many random translation/rotation insertion guesses the Subdocker makes per BatchDocker queue limit.
• --structs_per_loading: Maximum number of successful geometries to export out of all validated matches, per conformer tested.
• --num_clusters & --min_radius: Settings for the VoronoiClustering sampler that determine the density and minimum pore volume of chosen docking nodes within the material.
• --remove_species: Pre-cleans the CIF file of specified elements (like free solvent) before docking.

Constraints

• Environment: Requires atomistic-agent conda environment where VOID, rdkit, and pymatgen are accessible.
• Loading Size: By default, this script handles single-guest loadings per unit cell. Heavy multiple guest loading (--max_loading > 1) may scale exponentially in computational time depending on pore size.
• Outputs: Everything is standardized to CIF files for compatibility with subsequent DFT or MLIP workflows.

References

1. VOID Library
2. Pymatgen pymatgen.core.Structure and pymatgen.core.Molecule
3. RDKit cheminformatics (MMFF94 structural optimization)

---

Author: Mingrou Xie Contact: GitHub @mingrouxie