diffdock

Predict 3D binding poses for protein-ligand complexes using diffusion models by providing PDB structures and SMILES strings as inputs. The process generates ranked binding poses with confidence scores, accelerating structure-based drug design and virtual screening workflows.

Yes, you can perform batch virtual screening and ensemble docking across multiple protein conformations. The system supports processing multiple ligands simultaneously and docking against several protein structures to evaluate binding robustness and generate confidence rankings.

Supported input formats for molecular docking include PDB files for protein structures and SMILES, SDF, or MOL2 files for ligand descriptions. Protein sequences can also be provided, which are processed via ESMFold to generate the necessary structures.

Yes, diffusion-based ligand docking requires a Python-based environment with dependencies like pandas and rdkit. It utilizes provided inference scripts and YAML configuration templates to execute the pose prediction and virtual screening workflows.

Confidence scoring for pose prediction in structure-based drug design works by ranking generated ligand binding poses. The diffusion model outputs multiple poses per ligand, assigning a confidence score to each, allowing researchers to filter and prioritize the most reliable structural hypotheses.

Yes, predicted protein-ligand poses can be integrated with downstream scoring and visual inspection workflows like GNINA and MM/GBSA. This allows researchers to further refine the generated pose hypotheses and evaluate binding affinities for drug discovery.