diffdock

Generate protein-ligand binding poses with confidence scores via diffusion.

1|Updated Mar 19, 2026

Search Tags:#docking #virtual-screening #protein-ligand #pose-prediction #diffdock #confidence-scores

Ready to run

npx skills add https://github.com/JosephWoodall/noosphere --skill diffdock-josephwoodall

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: diffdock
Source: https://github.com/JosephWoodall/noosphere/tree/main/.agent/skills/diffdock
Command: npx skills add https://github.com/JosephWoodall/noosphere --skill diffdock-josephwoodall

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires pandas, rdkit, Bio, pytorch_lightning, yaml, numpy, scipy, esm, torch, and includes scripts (resource) and references (resource) and assets (resource) components.

What problem does it solve?

DiffDock generates protein-ligand binding poses with accompanying confidence scores.

Core Features & Use Cases

Support single docking, batch docking, and ensemble docking (including sequence-based targets via ESMFold) for structure-based drug design.
Predict 3D binding poses from PDB or SMILES with per-pose confidence scores for screening and design.
Integrate with downstream rescoring tools and provide utility scripts for batch CSV preparation, results analysis, and environment validation.

Quick Start

Run a GPU-enabled docking job by providing a protein (PDB or sequence) and a ligand (SMILES or file) to generate and inspect the top poses with confidence scores.

Frequently Asked Questions about diffdock

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

FAQPage Schema

How do I predict protein-ligand binding poses from a PDB file and SMILES string?▼

Protein-ligand binding pose prediction is done by providing a protein PDB structure and a ligand SMILES string. DiffDock uses diffusion-based generation to output 3D binding poses with accompanying confidence scores for screening and design.

Can I run batch virtual screening for multiple ligands against a single protein target?▼

Batch virtual screening is supported by preparing a CSV file containing multiple ligand SMILES strings. The batch docking workflow processes the ligands against a single protein target and generates top poses with confidence scores for structure-based drug design.

How does sequence-based ensemble docking work when I only have a protein sequence?▼

Sequence-based ensemble docking works by using ESMFold to generate a 3D protein structure directly from the input sequence. The workflow then proceeds with diffusion-based pose generation to predict ligand binding poses and confidence scores.

Do I need a GPU to run diffusion-based pose generation for structure-based drug design?▼

A GPU-enabled environment is required to run diffusion-based pose generation efficiently. The workflow relies on PyTorch Lightning and Torch to execute the computational diffusion models and generate 3D binding poses with confidence scores.

What's the best way to analyze confidence scores from generated docking poses?▼

Analyzing confidence scores from generated docking poses is done using provided utility scripts for results analysis. The scripts process the output files to evaluate per-pose confidence scores, enabling you to filter and rank candidates for downstream rescoring.