diffdock

Predict 3D protein-ligand binding poses with per-pose confidence scores.

1|2|Updated Apr 29, 2026

Search Tags:#protein-ligand #molecular-docking #rdkit #pose-prediction #diffusion-models #esmfold #diffdock

Ready to run

npx skills add https://github.com/fuzzy-dynamics/strings --skill diffdock-fuzzy-dynamics

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: diffdock
Source: https://github.com/fuzzy-dynamics/strings/tree/main/packages/skills/diffdock
Command: npx skills add https://github.com/fuzzy-dynamics/strings --skill diffdock-fuzzy-dynamics

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires pandas, rdkit, and includes scripts (resource) and references (resource) and assets (resource) components.

What problem does it solve?

DiffDock provides diffusion-model based docking to generate plausible 3D protein-ligand poses and accompanying confidence estimates, enabling structure-based drug design workflows without manual docking trials.

Core Features & Use Cases

Predicts 3D binding poses for small molecules with protein targets.
Generates per-pose confidence scores to enable robust ranking and downstream scoring.
Supports single complex docking, batch docking, and virtual screening; inputs include PDB structures or sequences (via ESMFold) and ligands as SMILES or structure files (SDF/MOL2).
Enables ensemble docking for protein flexibility and straightforward integration with downstream scoring pipelines.
Use Case: Screen a library of SMILES against a target protein to prioritize top poses for experimental validation.

Quick Start

Run inference with the default configuration to generate poses and review the top predictions.

Frequently Asked Questions about diffdock

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I predict 3D protein-ligand binding poses from a SMILES string and a protein sequence?▼

Predict 3D protein-ligand binding poses by inputting protein sequences via ESMFold and ligands as SMILES strings. DiffDock generates plausible 3D binding poses with accompanying per-pose confidence scores for ranking.

Can I run batch molecular docking and virtual screening across a library of small molecules?▼

Batch molecular docking and virtual screening are supported natively. You can screen a library of SMILES against a target protein to prioritize top poses for experimental validation using generated confidence scores.

What is diffusion-model based docking and how does it estimate binding confidence?▼

Diffusion-model based docking generates plausible 3D protein-ligand binding poses by iteratively refining random noise. It outputs per-pose confidence scores to enable robust ranking and integration with downstream scoring pipelines.

Does DiffDock work with PDB structures and SDF or MOL2 ligand files?▼

DiffDock works with PDB structures or sequences for protein targets, and accepts ligands as SMILES or structure files including SDF and MOL2. This flexibility supports various structure-based drug design workflows.

How do I handle protein flexibility during molecular docking?▼

Handle protein flexibility during molecular docking by utilizing ensemble docking capabilities. This approach generates multiple plausible 3D binding poses to account for target conformational variations.

Do I need RDKit and pandas to run pose prediction workflows?▼

RDKit and pandas are required dependencies for running pose prediction workflows. They handle molecular data parsing and dataframe operations to support DiffDock's configurable inference and batch processing.