diffdock

Name: diffdock
Availability: InStock
Author: estrella-231

Predict ranked protein–ligand binding poses from PDB or SMILES inputs.

Updated May 24, 2026

Search Tags:#virtual screening #molecular docking #diffusion models #confidence scoring #protein-ligand #pose prediction #structure-based drug design

Ready to run

npx skills add https://github.com/Estrella-231/Mathematical_modeling_tongmeng --skill diffdock-estrella-231

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: diffdock
Source: https://github.com/Estrella-231/Mathematical_modeling_tongmeng/tree/main/.agents/skills/diffdock
Command: npx skills add https://github.com/Estrella-231/Mathematical_modeling_tongmeng --skill diffdock-estrella-231

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires os, pathlib, re, argparse, json, collections, and includes scripts (resource) and references (resource) and assets (resource) components.

What problem does it solve?

DiffDock helps you predict how a small molecule binds by generating plausible 3D protein–ligand binding poses and associated confidence scores, so you can prioritize candidates for downstream scoring and validation.

Core Features & Use Cases

Pose prediction (not affinity): generates ranked 3D docking poses plus confidence to estimate prediction certainty.
Flexible input support: accepts proteins from PDB or protein sequences (via ESMFold) and ligands from SMILES or structure files (SDF/MOL2).
Batch virtual screening: runs many complexes via a CSV workflow and provides confidence-based result analysis.

Quick Start

Run a single docking by providing a protein PDB and a ligand SMILES, saving ranked 3D poses and confidence scores to your results folder.

Frequently Asked Questions about diffdock

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

FAQPage Schema

How does diffusion-based molecular docking predict protein-ligand binding poses?▼

Diffusion-based molecular docking predicts protein-ligand binding poses by generating plausible 3D structures from PDB or sequence inputs and SMILES ligands, providing confidence-scored rankings for virtual screening.

Can I run batch virtual screening for multiple protein-ligand complexes?▼

Yes, you can run batch virtual screening by providing multiple protein-ligand complexes via a CSV workflow, which generates confidence-scored pose predictions to help prioritize candidates across many targets.

What input formats are supported for protein and ligand structure prediction?▼

Structure prediction accepts protein inputs as PDB files or raw protein sequences via ESMFold, and ligand inputs as SMILES strings or structural file formats including SDF and MOL2.

Does the confidence score from diffusion docking indicate binding affinity?▼

No, the confidence score indicates pose prediction reliability rather than binding affinity, meaning it estimates the certainty of the generated 3D binding pose for downstream validation.

How do I predict binding poses when I only have a protein sequence instead of a PDB structure?▼

To predict binding poses with only a protein sequence, the diffusion docking pipeline utilizes ESMFold to predict the protein structure first, then proceeds to generate ligand binding poses.

What's the best way to prioritize candidates after structure-based drug design screening?▼

The best way to prioritize candidates after structure-based screening is to rank the generated 3D docking poses using their associated confidence scores, prioritizing high-confidence predictions for downstream scoring.