Terminal
Terminal/Skill
Guides molecular docking with AutoDock Vina for drug discovery, covering receptor and ligand prep, grid boxes, and virtual screening.
By Amazon Web Services
Works with
npx skills add awslabs/hcls-agent-skills --skill molecular-dockingAdds decision logic for receptor/ligand preparation tool selection, grid box sizing, exhaustiveness tuning, and validation protocol. Focuses on gotchas that produce silently wrong poses.
Receptor preparation tool:
├── Quick exploration / single ligand → obabel (-d, reduce, -xr)
└── Production virtual screen → prepare_receptor (ADFR) or meeko
└── Handles altlocs, metals, non-standard residues correctly
Ligand preparation:
├── Single ligand from SMILES/SDF → obabel --gen3d -h
├── Large library (>100 ligands) → meeko (mk_prepare_ligand.py)
└── Stereocenters undefined? → Enumerate explicitly BEFORE docking
Grid box placement:
├── Co-crystal ligand available → Center on ligand centroid
└── No co-crystal → Use pocket predictor (fpocket, P2Rank, SiteMap)
Grid box sizing:
└── Ligand longest diameter + 10–15 Å padding per axis
└── Typical: 20–30 Å per side for drug-like molecules
Exhaustiveness:
├── Quick exploration → 8 (Vina default)
├── Production / ranking → 32
└── Large/flexible ligands (>8 rotatable bonds) → 64+
Validation gate (MUST pass before trusting screen results): Redock co-crystal ligand → RMSD < 2 Å against crystallographic pose. If fails, fix preparation or grid before screening.
| Parameter | Recommended | When to change |
|---|---|---|
exhaustiveness | 32 | 64+ for >8 rotatable bonds |
num_modes | 10 | Increase for ensemble analysis |
energy_range | 3 kcal/mol | Widen to capture diverse poses |
| Box padding | +10–15 Å | Larger for allosteric sites |
| Protonation pH | 7.0 (reduce default) | Use PROPKA/H++ for acidic pockets (e.g., aspartyl proteases pH 4.5) |
Score interpretation:
Wrong: Leaving crystallographic waters in the receptor Right: Strip HOH/WAT residues before PDBQT conversion (unless explicit-water docking) Why: Waters block the binding site and produce unphysical poses
Wrong: Using default pH 7 protonation for all binding sites Right: Use PROPKA or H++ for acidic/basic pockets Why: Wrong protonation alters H-bond networks and produces incorrect poses
Wrong: Grid box barely enclosing the binding site Right: Ligand diameter + 10–15 Å padding per axis Why: Under-padding clips the site and biases poses toward box center
Wrong: Default exhaustiveness = 8 for large/flexible ligands
Right: Use 32+ for production; 64+ for >8 rotatable bonds
Why: Low exhaustiveness misses global minima, unreliable rankings
Wrong: Skipping redocking validation before screening Right: Redock co-crystal ligand, verify RMSD < 2 Å Why: RMSD ≥ 2 Å means preparation or grid is wrong; screen results untrustworthy
Wrong: Using obabel -xr PDBQT for production virtual screens
Right: Use prepare_receptor (ADFR) or meeko for production
Why: obabel may miss altlocs, metals, or non-standard residues
Wrong: Running obabel --gen3d on SMILES with undefined stereocenters
Right: Provide isomeric SMILES or enumerate stereoisomers explicitly
Why: Undefined stereocenters resolved arbitrarily — may dock wrong enantiomer
Wrong: Comparing Vina affinities across different receptors or grid configs Right: Only compare within same receptor + grid; re-rank when changing either Why: Scores are context-dependent, not transferable