Multi-Modal Stucture Prediction with Chai-1

Chai-1 is a general-purpose deep learning model for biomolecular structure prediction, designed to work across a wide range of inputs: proteins, nucleic acids, small-molecule ligands, and multi-chain assemblies. Unlike models specialized for a single domain, Chai-1 supports multiple modalities in one architecture and can be optionally guided by experimental constraints such as cross-linking mass spectrometry, epitope mapping, or known contact residues.

Because Chai-1 can run in both full multi-sequence-alignment (MSA) mode and single-sequence mode, it adapts well to different stages of discovery—from high-throughput screening against well-characterized targets to exploratory modeling of novel or variable sequences where evolutionary data is sparse.

Method Overview

Chai-1 follows a paired-attention architecture similar in spirit to AlphaFold3, but with key differences. Chai-1 can incorporate residue-level embeddings from protein language models, enabling strong single-sequence performance. Chai-1 can also incorporate optional pocket, contact, and docking restraints to bias the model's predictions in the direction of experimental priors. The full architecture diagram is shown here:

Chai-1 architecture diagram

The architecture of Chai-1 (Figure 1 from the preprint).

During inference, Chai-1 can sample multiple structures (via trunk and diffusion sampling) and ranks them by calibrated confidence metrics such as interface pTM (ipTM) (Figure 2 from the preprint). This is only a brief summary of how Chai-1 works; for full details, see the preprint.

Performance

On the PoseBusters protein–ligand benchmark, restricted to post-cutoff structures, Chai-1 achieves a 77% success rate (ligand RMSD ≤ 2 Å) from sequence + SMILES alone, comparable to AlphaFold3's reported 76%. When provided the apo protein structure as a prompt, success rises to 81%.

PoseBusters benchmark

Protein–ligand pose prediction success rates (PoseBusters V1) and protein accuracy (DockQ and RMSD). See the preprint for full details.

For protein–protein complexes in a low-homology set, Chai-1 with MSAs outperforms AlphaFold-Multimer 2.3 (0.751 vs 0.677 DockQ success). In single-sequence mode without MSAs, it still matches or exceeds AF2.3 on many interface types, particularly antibody–protein complexes, where evolutionary signal is often weak.

Chai-1 also supports nucleic acid modeling without MSAs, achieving accuracy comparable to RoseTTAFold2NA despite lacking evolutionary inputs for RNA/DNA.

Nevertheless, a word of caution is warranted. Benchmarks from Peter Škrinjar and co-workers demonstrate that the performance of current protein–ligand co-folding methods, including Chai-1, depends largely on similarity to the training set. This suggests that particular caution is warranted for novel or poorly explored regions of biochemical space.

Protein-ligand induced fit example

Similarity to the training set predicts cofolding performance.

(On Practical Cheminformatics, Pat Walters discusses this paper and a variety of other studies demonstrating that Chai-1 and related methods don't yet understand binding physics correctly.)

Applications

Chai-1's breadth and ability to run without MSAs makes it suitable for ligand pose generation, protein–protein docking, antibody–antigen modeling, nucleic-acid-complex prediction without the need for specialized MSAs, or simply exploratory protein-structure modeling.

How does this work in practice? Here's what Pat Walters has to say about the current use of co-folding methods in drug discovery:

I have been using co-folding methods in a similar way to how I use molecule generation techniques— as tools to develop hypotheses that can be tested experimentally. We have many methods that tell us if a molecule binds, such as biochemical assays, size exclusion chromatography-MS, examining binding in a DEL screen, or performing various other experiments. However, apart from X-ray crystallography or cryoEM, we have very few experiments that can definitively identify where a compound binds. In most cases, we need a hypothesis to guide experiments, such as mutagenesis or installing a photoaffinity label. Even more definitive experiments, like NMR or HDX-MS, are much more useful when a binding hypothesis is available. Co-folding provides an easy way to generate these hypotheses; then, it's up to our scientific knowledge, intuition, and creativity to decide how to proceed.

This conclusion—that co-folding methods are promising hypothesis generators but no substitute for experimental data—seems to match what we're seeing from lots of other drug-discovery teams.

Accessing Chai-1 in Rowan

Rowan provides browser-based access to Chai-1, with support for all major input types. Users can run Chai-1 in:

Full mode (with MSA search) for maximum accuracy.
Single-sequence mode for faster turnaround or novel sequences.

An interactive 3D viewer lets you inspect predicted contacts, ligand geometry, and inter-chain interfaces.

Here's what the result of a Chai-1 calculation looks like on Rowan:

Chai-1 example

View this calculation on Rowan.

Python API Access

Chai-1 calculations can also easily be submitted via Rowan's Python API, which is available for free to all users:

import rowan

rowan.api_key = "your-rowan-api-key"

workflow = rowan.submit_protein_cofolding_workflow(
    initial_protein_sequences=[
        "MENFQKVEKIGEGTYGVVYKARNKLTGEVVALKKIRLDTETEGVPSTAIREISLLKELNHPNIVKLLDVIHTENKLYLVFEFLHQDLKKFMDASALTGIPLPLIKSYLFQLLQGLAFCHSHRVLHRDLKPQNLLINTEGAIKLADFGLARAFGVPVRTYTHEVVTLWYRAPEILLGCKYYSTAVDIWSLGCIFAEMVTRRALFPGDSEIDQLFRIFRTLGTPDEVVWPGVTSMPDYKPSFPKWARQDFSKVVPPLDEDGRSLLSQMLHYDPNKRISAKAALAHPFFQDVTKPVPHLRL"
    ],
    initial_smiles_list=["CCC(C)CN=C1NCC2(CCCOC2)CN1"],
    name="Chai-1 cofolding job",
    model="chai_1r",
)

workflow.wait_for_result()
workflow.fetch_latest(in_place=True)

print(workflow.data)

This prints the following data; the corresponding 3D structure can be viewed through Rowan's web interface.

{
  'lddt': [0.895, 0.938, 0.967, 0.976, 0.971, 0.971, 0.971, 0.97, 0.963, 0.94, 0.908, 0.873, 0.835, 0.838, 0.836, 0.874, 0.921, 0.945, 0.968, 0.977, 0.982, 0.981, 0.977, 0.97, 0.948, 0.961, 0.968, 0.973, 0.982, 0.982, 0.976, 0.964, 0.941, 0.921, 0.889, 0.771, 0.72, 0.691, 0.643, 0.678, 0.716, 0.742, 0.784, 0.832, 0.846, 0.881, 0.885, 0.892, 0.895, 0.917, 0.909, 0.931, 0.924, 0.907, 0.937, 0.942, 0.94, 0.95, 0.967, 0.984, 0.987, 0.988, 0.984, 0.981, 0.973, 0.967, 0.968, 0.958, 0.951, 0.938, 0.89, 0.862, 0.796, 0.735, 0.837, 0.902, 0.942, 0.953, 0.969, 0.974, 0.978, 0.982, 0.98, 0.965, 0.98, 0.982, 0.988, 0.982, 0.98, 0.987, 0.987, 0.984, 0.981, 0.982, 0.977, 0.974, 0.969, 0.984, 0.987, 0.987, 0.987, 0.989, 0.989, 0.99, 0.99, 0.99, 0.99, 0.99, 0.99, 0.99, 0.989, 0.99, 0.989, 0.986, 0.986, 0.987, 0.986, 0.975, 0.978, 0.978, 0.967, 0.951, 0.956, 0.964, 0.957, 0.942, 0.946, 0.969, 0.973, 0.983, 0.96, 0.961, 0.986, 0.985, 0.987, 0.984, 0.984, 0.983, 0.988, 0.989, 0.989, 0.988, 0.985, 0.972, 0.916, 0.881, 0.83, 0.86, 0.866, 0.869, 0.84, 0.826, 0.778, 0.791, 0.778, 0.796, 0.845, 0.804, 0.759, 0.784, 0.811, 0.835, 0.843, 0.859, 0.925, 0.969, 0.983, 0.981, 0.98, 0.985, 0.988, 0.987, 0.979, 0.983, 0.987, 0.982, 0.97, 0.939, 0.949, 0.909, 0.978, 0.983, 0.987, 0.983, 0.984, 0.989, 0.989, 0.987, 0.99, 0.99, 0.99, 0.989, 0.99, 0.99, 0.989, 0.989, 0.989, 0.988, 0.987, 0.986, 0.988, 0.989, 0.989, 0.986, 0.979, 0.958, 0.968, 0.969, 0.986, 0.987, 0.987, 0.989, 0.99, 0.989, 0.99, 0.99, 0.989, 0.989, 0.989, 0.989, 0.99, 0.989, 0.989, 0.989, 0.988, 0.988, 0.989, 0.988, 0.987, 0.988, 0.989, 0.987, 0.987, 0.987, 0.988, 0.989, 0.988, 0.988, 0.988, 0.989, 0.989, 0.989, 0.989, 0.988, 0.987, 0.985, 0.964, 0.97, 0.985, 0.985, 0.985, 0.988, 0.978, 0.966, 0.986, 0.986, 0.987, 0.989, 0.989, 0.989, 0.989, 0.99, 0.989, 0.988, 0.99, 0.99, 0.99, 0.989, 0.99, 0.989, 0.989, 0.987, 0.988, 0.989, 0.99, 0.989, 0.99, 0.989, 0.99, 0.99, 0.99, 0.989, 0.99, 0.989, 0.99, 0.99, 0.988, 0.988, 0.989, 0.989, 0.989, 0.989, 0.985, 0.985, 0.971, 0.971, 0.939, 0.942], 
  'model': 'chai_1r', 
  'scores': {'ptm': 0.97, 'iptm': 0.796, 'avg_lddt': 0.926, 'confidence_score': 0.831}, 
  'messages': [], 
  'affinity_score': None, 
  'use_msa_server': True, 
  'use_potentials': False, 
  'pocket_constraints': [], 
  'contact_constraints': [], 
  'initial_smiles_list': ['CCC(C)CN=C1NCC2(CCCOC2)CN1'], 
  'use_templates_server': False, 
  'predicted_structure_uuid': '0d3bc244-e7f2-473d-bd4a-439b1e03a2e8', 
  'initial_protein_sequences': ['MENFQKVEKIGEGTYGVVYKARNKLTGEVVALKKIRLDTETEGVPSTAIREISLLKELNHPNIVKLLDVIHTENKLYLVFEFLHQDLKKFMDASALTGIPLPLIKSYLFQLLQGLAFCHSHRVLHRDLKPQNLLINTEGAIKLADFGLARAFGVPVRTYTHEVVTLWYRAPEILLGCKYYSTAVDIWSLGCIFAEMVTRRALFPGDSEIDQLFRIFRTLGTPDEVVWPGVTSMPDYKPSFPKWARQDFSKVVPPLDEDGRSLLSQMLHYDPNKRISAKAALAHPFFQDVTKPVPHLRL'], 
  'ligand_binding_affinity_index': None
}

If you're interested in trying out Chai-1 today, make an account for free and start running calculations in minutes!