chai1

1# Chai-1
2 
3Chai-1 is an all-atom diffusion co-folder in the same family as Boltz-2 and
4AlphaFold3: a multi-entity FASTA in, mmCIF plus pTM/ipTM/pLDDT out, with
5protein, RNA, DNA, and SMILES-ligand chains all first-class. It and `boltz`
6cover the same surface; running both and keeping designs that pass either is a
7common consensus filter, and Chai's Python entry point makes it the easier of
8the two to embed in a loop. Code and weights are Apache-2.0 — commercial use
9including drug discovery is explicitly permitted
10(github.com/chaidiscovery/chai-lab).
11 
12## Running it
13 
14```python
15from pathlib import Path
16from chai_lab.chai1 import run_inference
17 
18Path("complex.fasta").write_text("""
19>protein|name=target
20MVTPEGNVSLVDESLLVGVTDEDRAVRS...
21>protein|name=binder
22AIQRTPKIQVYSRHPAENG...
23>ligand|name=cofactor
24CCCCCCCCCCCCCC(=O)O
25""".strip())
26 
27candidates = run_inference(
28    fasta_file=Path("complex.fasta"),
29    output_dir=Path("out/"),
30    num_trunk_recycles=3,
31    num_diffn_timesteps=200,
32    seed=42,
33    device="cuda:0",
34    use_esm_embeddings=True,
35)
36print([rd.aggregate_score.item() for rd in candidates.ranking_data])
37```
38 
39The FASTA header is `>{entity_type}|name={id}` with `entity_type` ∈
40{`protein`, `rna`, `dna`, `ligand`}; ligand records carry a SMILES string as
41the sequence body, and modified residues are written inline as
42`...AAK(SEP)AAG...`. From the shell the same job is `chai-lab fold
43complex.fasta out/ --use-msa-server`. Without `--use-msa-server` (or
44…