diffdock

To predict protein-ligand binding poses, you provide a protein PDB file or sequence and a ligand SMILES string. The diffusion-based inference generates ranked 3D binding poses output as SDF files alongside per-pose confidence scores.

Yes, you can run batch virtual screening by providing a CSV file containing multiple ligand inputs. This allows you to process multiple protein-ligand complexes sequentially and generate ranked 3D poses with confidence scores for each entry.

No, DiffDock does not predict binding affinity scores. It generates ranked 3D ligand binding poses with confidence scores to indicate pose reliability, which helps you triage which complexes need further downstream scoring and refinement.

You do not need a pre-existing protein structure file. You can provide a protein sequence instead of a PDB file, and the skill utilizes ESFold to generate the required structural input for diffusion docking pose prediction.

Diffusion docking is a structure-based drug design technique that uses diffusion-model inference to predict 3D binding poses. It accelerates virtual screening workflows by automating pose generation and outputting ranked structural hypotheses without manual docking setup.

Predicted ligand poses need downstream scoring and refinement because the skill only provides confidence scores for pose reliability, not exact binding affinities. You must use additional tools to accurately calculate binding energy and refine the molecular interactions.