boltz

1# Boltz-2
2 
3Boltz-2 is the open-weights diffusion co-folder closest in surface to
4AlphaFold3: a YAML describing protein, DNA, RNA, and ligand chains in, mmCIF
5plus pTM/ipTM/pLDDT confidences out, with an optional small-molecule affinity
6head. Among our four co-fold skills it is the default for binder-validation
7campaigns — fully open MIT weights and the fastest sampler; pick `chai1` when
8you want a second independent model for consensus, `openfold3` when AF3-faithful
9settings matter, and `esmfold2` when you can live without an MSA. Code and
10weights are MIT (PyPI `boltz`, github.com/jwohlwend/boltz).
11 
12## Running it
13 
14```yaml
15# complex.yaml
16version: 1
17sequences:
18  - protein:
19      id: A
20      sequence: MVTPEGNVSLVDESLLVGVTDEDRAVRS... # target
21  - protein:
22      id: B
23      sequence: AIQRTPKIQVYSRHPAENG... # binder
24  - ligand:
25      id: L
26      smiles: 'N[C@@H](Cc1ccc(O)cc1)C(=O)O' # or  ccd: SAH
27```
28 
29```bash
30boltz predict complex.yaml \
31    --use_msa_server --out_dir out/ --recycling_steps 3 --diffusion_samples 5
32```
33 
34Each protein chain needs an MSA; without one the run exits before the model
35loads. `--use_msa_server` queries `api.colabfold.com` (expect a 30–90 s pause
36per chain) and is the right default unless you already have an `.a3m` to name
37under `msa:` in the YAML. Setting `msa: empty` forces single-sequence mode —
38that is an accuracy sacrifice, not a speed or memory optimization, because the
39MSA search runs on CPU before the GPU stage starts.
40…