Use Chai-1 Online
Commercially Available Chai-1 No-Code Web Server
Chai-1: Multi-Modal Foundation Model for Biomolecular Structure Prediction
Chai-1 is a multi-modal foundation model for predicting the 3D structures of proteins, small molecules and nucleic acid complexes. It was developed by Jacques Boitreaud, Jack Dent, Matthew McPartlon, Joshua Meier, Vinicius Reis, Alex Rogozhnikov and Kevin Wu at Chai Discovery and released as a bioRxiv preprint. The code and weights are on GitHub for non-commercial use, and the authors also provide a web server.
Structure-based drug discovery needs models that handle protein-ligand and protein-protein complexes, not only single chains. Chai-1 largely follows the AlphaFold3 architecture, adds embeddings from a protein language model, and accepts optional experimental restraints, so predictions can be steered by information such as epitope residues or cross-linking contacts.
Property | Detail |
|---|---|
Inputs | FASTA and SMILES; optional MSAs, templates, language model embeddings and restraints |
Outputs | 25 predicted structures per run, ranked by a confidence model |
Restraint types | Pocket, contact and docking |
Limit | Up to 2048 tokens |
Venue | bioRxiv preprint, DOI 10.1101/2024.10.10.615955 |
How Chai-1 Predicts Protein-Ligand and Multimer Structures with Optional Restraints
Inputs: sequences and ligand SMILES, plus optional MSAs and templates.
Language model track: per-residue embeddings from a 3-billion-parameter protein language model are added, which enables single-sequence mode.
Restraints: pocket, contact and docking features let users supply known interface information.
Sampling and ranking: 4 recycles, 5 trunk samples and 5 diffusion samples yield 25 structures that a confidence model ranks.
Chai-1 Benchmark Results for Ligands, Antibodies and Multimers
Benchmark | Result |
|---|---|
PoseBusters (427 structures), ligand RMSD under 2 Angstroms | Chai-1 77% vs AlphaFold3 76% (reported values) |
PoseBusters with apo protein structure | 81% |
Multimers, mean DockQ success | 0.751 vs 0.677 for AlphaFold-Multimer 2.3 |
Multimers, single-sequence mode | 0.698 |
Antibody-antigen DockQ-acceptable rate | 35% baseline, 57% with one contact restraint |
CASP15 monomers, mean LDDT | 0.849 vs 0.843 |
The authors report that restraints boost performance by double-digit percentage points, and that single-sequence mode retains most performance without MSAs. In one example (PDB 7SYV), DockQ rose from 0.10 to 0.81 with epitope restraints.
What is Tamarind Bio?
Tamarind Bio is a no-code bioinformatics platform built to give life scientists and researchers access to powerful computational tools. Many cutting-edge machine learning models are hard to deploy and use. Tamarind provides an intuitive, web-based environment that removes the complexity of high-performance computing, software dependencies and command-line interfaces.
The platform is designed for biologists, chemists and other researchers who may not have a background in programming or cloud infrastructure but want to run models on their own data. Key features include:
A user-friendly graphical interface for setting up and launching experiments
A robust API for integration into existing research pipelines
An automated system for managing and scaling computational resources
Tamarind treats information and data security as a top priority, as detailed in its Trust Center and Terms of Service.
Accelerating Discovery with Chai-1 on Tamarind Bio
Protein-ligand pose prediction: predict how small molecules bind, with or without a known apo structure.
Protein complex modeling: predict multimer and antibody-antigen interfaces.
Restraint-guided prediction: incorporate epitope mapping or cross-linking mass spectrometry information.
Fast single-sequence runs: predict structures when MSAs are unavailable.
How to Use Chai-1 on Tamarind Bio
Open the tool: log in to tamarind.bio and open Chai-1.
Provide sequences: enter protein, DNA or RNA sequences for each chain.
Add ligands: enter small molecules as SMILES, if exposed.
Choose MSA mode: use MSAs or single-sequence mode, if exposed.
Add restraints: supply pocket, contact or docking restraints, if exposed.
Run the job: submit and wait for the prediction.
Download and rank: retrieve the structures and review confidence scores to pick the best.
Use downstream: inspect interfaces, run docking refinement or design follow-up.
Things to Keep in Mind
Chains may be predicted correctly but placed in the wrong relative orientation, especially without contact information.
Predictions can change substantially when modified residues are replaced with standard analogs.
The model sometimes fails to respect input ligand chirality.
High-quality antibody-antigen predictions remain rare, even with restraints.
Nucleic acid MSAs are not used, and inference is capped at 2048 tokens.
The authors are employees of Chai Discovery, and AlphaFold3 comparisons rely on reported values.
Source: Boitreaud, J., Dent, J., McPartlon, M., Meier, J., Reis, V., Rogozhnikov, A. and Wu, K., Chai-1: Decoding the molecular interactions of life. bioRxiv preprint (v2, 2024), DOI: 10.1101/2024.10.10.615955. Code: https://github.com/chaidiscovery/chai-lab/