Docking proposes interaction geometries. Screening ranks a library for further investigation. Neither a favorable docking score nor an attractive pose establishes measured binding.
Choose the docking problem
Distinguish protein-ligand docking, peptide-protein docking, and protein-protein assembly. Determine whether the binding site is known, whether the receptor state is appropriate, and what flexibility the method actually models.
| Setup | Tools to investigate | Important distinction |
|---|---|---|
| Protein-ligand pose search and scoring | GNINA, AutoDock Vina | Search space, receptor assumptions, and native scoring |
| Blind ligand docking | DiffDock-L | Learned pose generation and confidence, not guaranteed site identification |
| Defined-pocket ligand or peptide docking | PocketXMol Dock | Pocket definition and ligand/peptide mode |
| Protein-protein assembly | LightDock or appropriate complex prediction tools | Interface evidence, chain identity, and domain flexibility |
Prepare receptor and partners
Choose the relevant biological assembly, receptor conformation, protonation assumptions, cofactors, and ligand stereochemistry. A receptor taken directly from a crystal file may include irrelevant copies or omit essential atoms.
Use repair and validation where appropriate. Define a search box in the units and convention requested by the service. Preserve coordinate frames when the setup uses a reference ligand or known pocket center.
Review poses before ranking
Inspect pocket occupancy, clashes, strained conformations, buried groups, and plausible contacts in the viewer. Compare pose families rather than selecting one isolated score without context.
Keep docking energy, learned pose confidence, and affinity-style scores separate. Different engines’ numerical outputs are not interchangeable. Interaction Fingerprint can help compare contact patterns among compatible complexes.
Scale to a screening library
Pilot a representative subset, preserve compounds that fail preparation, and declare how you select poses and aggregate scores. Review diversity and ADMET risks before choosing a shortlist.
Use bulk submission rules to distinguish one receptor against N ligands from all combinations. Advance selected hypotheses to compatible simulation, rescoring, or experiments; do not interpret a screening rank as a measured dissociation constant.