Rowan Docs

Docking

Python API · stjames models · API example · API example

How it works

Docking predicts plausible binding poses for ligands in a protein pocket. Rowan docks prepared ligand conformations, removes duplicate poses, and optionally refines them. Results include pose structures, protein–ligand complexes, docking scores, and available strain, MM/GBSA, and geometry checks.

Settings

  • Pocket: set the search box manually, use a saved or docked-ligand pocket, or select "Blind Dock" to search across the protein. Larger search regions need more sampling.
  • Docking settings: choose AutoDock Vina, QuickVina, or GNINA. "Scoring function" selects how poses are ranked. QuickVina supports Vina scoring only.
  • Exhaustiveness and Max poses: increase exhaustiveness for a more thorough, slower search. The pose limit applies per starting conformer, so conformer search can produce more total poses.
  • Run conformer search? and Optimize ligand(s)?: generate multiple starting geometries or relax the supplied geometry. Conformer search automatically requires optimization.
  • Local opt. and MM/GBSA?: minimize noncovalent poses and estimate interaction energies. With conformer search and optimization enabled, also calculate ligand strain.
  • Induced-fit docking?: allow the receptor to adapt to candidate poses; requires AutoDock Vina or QuickVina 2. "Max receptors" limits the sampled receptor ensemble; "Flexible sidechain radius" sets the distance from the ligand for selecting flexible residues, in Å.

Notes

Ligand preparation uses openconf for conformer generation, GFN2-xTB with ALPB water for optimization, and g-xTB with CPCM-X water for energy ranking. Without either preparation option, the supplied geometry is docked directly. Prepare the protein and choose appropriate protonation states before docking.

Ligand strain measures the energy cost of the bound conformation relative to the lowest-energy prepared conformer, after a restrained xTB relaxation. Without conformer search, this strain comparison is unavailable. If MM/GBSA refinement fails, Rowan retains raw poses and reports a warning. Covalent GNINA docking requires both reacting atom indices and Vina scoring; post-docking refinement and PoseBusters checks are skipped for these poses.

For noncovalent poses, PoseBusters checks chemical geometry and clashes with the receptor. Passing these checks does not establish binding. Lower Vina/Vinardo scores are better; higher GNINA CNN scores are better. Docking and MM/GBSA scores are approximate ranking tools, not measured binding affinities.

Induced fit uses soft docking, Boltz-2 co-folding, local receptor relaxation, and redocking in place of rigid docking. It adds runtime and ranks poses using docking score, receptor and ligand strain, and a penalty for failed geometry checks. MM/GBSA is reported separately. It requires receptor sequence records, removes waters, and rejects unsupported cofactors, ions, and residues. If no induced receptor yields a pose, the workflow fails.

Submission video

Further reading