Skip to main content
These are community-developed tools that complement AlphaFold 3. They are not officially supported by Google DeepMind.

JAAG: JSON Assembler for AlphaFold 3 (with Glycan Integration)

Overview

JAAG is a lightweight, web-based GUI tool that helps generate AlphaFold 3 input JSON files with integrated glycan support. It automates the creation of correct glycan syntax, reducing manual errors when preparing glycoprotein or glycan-protein complexes.

Key Features

  • Web-based interface: No installation required
  • Glycan support: Automated glycan syntax generation
  • CCD integration: Correct Chemical Component Dictionary codes
  • Bonded atom pairs: Automatic generation of bondedAtomPairs syntax
  • Error reduction: Validates input and prevents common mistakes
  • User-friendly: Intuitive GUI for non-programmers

Use Cases

  • Modeling glycoproteins
  • Creating glycan-protein complexes
  • Setting up covalent modifications
  • Learning AlphaFold 3 JSON format
  • Rapid prototyping of input files

Access and Resources

Compatibility

JAAG is compatible with standalone AlphaFold 3 installations only. It is not compatible with the AlphaFold 3 server at alphafoldserver.com.

Getting Started with JAAG

  1. Visit the JAAG web app
  2. Input your protein sequence
  3. Add glycan structures using the GUI
  4. Configure bonded atom pairs
  5. Export the generated JSON file
  6. Use the JSON file with your local AlphaFold 3 installation

Modeling Glycans with AlphaFold 3: Capabilities, Caveats, and Limitations

Overview

A comprehensive research paper and resource collection on modeling glycans (and other ligands) with AlphaFold 3. This work systematically modeled and assessed major glycan classes, providing practical guidance for users.

What’s Included

1. Step-by-Step Tutorial

  • Building ligand inputs from scratch
  • Applicable beyond glycans to other small molecules
  • Best practices for input preparation
  • Common pitfalls and how to avoid them

2. Ready-to-Run Scripts

  • Scripts for each major glycan class:
    • N-glycans
    • O-glycans
    • Glycosaminoglycans
    • Other glycan types
  • Copy-paste examples for quick start
  • Customizable templates

3. Comprehensive CCD Table

  • Complete table of CCD codes for all SNFG monosaccharides
  • Standard naming conventions
  • Chemical structures and properties
  • Bonding patterns

4. Analysis and Discussion

  • Caveats and limitations of AlphaFold 3 for glycan modeling
  • Accuracy assessment across glycan classes
  • When AlphaFold 3 works well vs. when it struggles
  • Validation strategies

5. Reproducible Data

  • Full AlphaFold 3 inputs archived on ModelArchive
  • Complete outputs for all modeled structures
  • Enables reproduction and comparison

When to Use This Resource

This resource is especially useful if your AlphaFold 3 ligand models appear stereochemically incorrect.
  • Glycan models look distorted
  • Unexpected bond angles or lengths
  • Clashes in glycan structures
  • Learning how to model complex carbohydrates
  • Understanding AlphaFold 3’s capabilities and limitations

Access and Resources

Key Takeaways from the Paper

  1. AlphaFold 3 can model many glycan types with good accuracy
  2. Careful input preparation is critical for success
  3. Validation is essential - don’t trust predictions blindly
  4. Some glycan classes are more challenging than others
  5. Stereochemistry must be checked carefully in outputs

Example Applications

  • Glycoproteins: Modeling antibodies with N-glycosylation
  • Lectins: Predicting lectin-glycan interactions
  • Glycan arrays: Screening multiple glycan structures
  • Drug discovery: Modeling glycan-binding proteins with inhibitors

Contributing Community Tools

Have you developed a tool for AlphaFold 3? Share it with the community!

How to Share Your Tool

  1. Publish on GitHub: Make your code openly available
  2. Write documentation: Clear usage instructions
  3. Contact the team: Email [email protected]
  4. Share in the community: Post on relevant forums and social media

Tool Development Guidelines

  • Open source: Use permissive licenses (MIT, Apache, etc.)
  • Well documented: README, examples, tutorials
  • Tested: Include test cases and validation
  • Compatible: Clearly state AlphaFold 3 version compatibility
  • Non-commercial: Respect AlphaFold 3’s non-commercial license

Useful Tool Categories

  • Input generation: Tools to create JSON files
  • Visualization: Structure viewers and analysis tools
  • Validation: Tools to assess prediction quality
  • Post-processing: Analysis of outputs
  • Integration: Connecting AlphaFold 3 with other software
  • Automation: Batch processing and workflows

Additional Resources

Official Resources

  • PyMOL: Visualization of predicted structures
  • ChimeraX: Advanced structure analysis and visualization
  • RDKit: Small molecule handling and validation
  • BioPython: Sequence manipulation and analysis

Databases

  • PDB: Protein Data Bank for template structures
  • CCD: Chemical Component Dictionary for ligand codes
  • UniProt: Protein sequences and annotations
  • SNFG: Symbol Nomenclature for Glycans

Disclaimer

Community tools are developed and maintained independently. Google DeepMind does not endorse or provide support for third-party tools.
When using community tools:
  • Verify compatibility with your AlphaFold 3 version
  • Test thoroughly before production use
  • Report issues to the tool developers
  • Cite both AlphaFold 3 and the community tool in publications
  • Review the tool’s license and terms of use

Build docs developers (and LLMs) love