diffdock

Predict protein-ligand binding poses using diffusion models by inputting a protein structure (PDB) or sequence and a ligand to generate 3D orientations with confidence scores. This approach automates structure-based discovery workflows for rapid candidate triage.

Yes, you can run batch molecular docking for a ligand library against a single target. The workflow supports batch-processing scripts with configurable inference parameters, enabling researchers to screen multiple ligands and obtain ranked binding poses for experimental validation.

Protein-ligand pose prediction works with just a protein sequence by leveraging ESMFold to generate the required structural conformation. This allows flexible inputs when a PDB file is unavailable, supporting both single and ensemble conformations for docking.

Integrate diffusion-based docking results with GNINA or MM/GBSA scoring workflows by exporting the predicted 3D ligand poses and confidence scores. This downstream compatibility allows for seamless affinity ranking and structural refinement in structure-based discovery pipelines.

Limitations of using diffusion models for molecular docking include the necessity of a PyTorch-based GPU environment with specific dependencies like torch-geometric and fair-esm. Additionally, generated poses require downstream scoring tools like GNINA or MM/GBSA for accurate affinity ranking and validation.

The best way to ensure reproducible results in batch molecular docking is to use the provided batch-processing scripts with configurable inference parameters and enforced environment checks. This end-to-end workflow standardizes the generation of protein-ligand binding poses and confidence metrics.