Structure prediction · Property prediction

Boltz-2

All-atom biomolecular structure and binding-affinity prediction.

Models complex structures and binding affinities, a critical component towards accurate molecular design. Boltz-2 is the first deep learning model to approach the accuracy of physics-based free-energy perturbation (FEP) methods, while running 1000x faster — making accurate in silico screening practical for early-stage drug discovery.

Documentation ↗ Input parameters ↗ Examples ↗ Home page ↗ Paper ↗ Source code ↗ License ↗

No jobs before this; it will run immediately.

Not logged in. Register an account to keep track of your jobs and results, and add presets and bookmarks. Email results to this address when complete:

Ready

Example preset

Choose an example to fill in its settings and input structures.

Input mode

Input

Job name Names the output directory for this run.

Molecules

Add one box per unique protein, ligand, DNA, or RNA entity. Set its native Boltz chain ID (or comma-separated IDs for identical copies), sequence or ligand, cyclic flag, and residue modifications. A protein's MSA may be left blank (generated by the MSA server when it is on, otherwise single-sequence), set to an absolute .a3m or .csv path, or set to empty for single-sequence mode.

Constraints (JSON, optional) Native Boltz constraints array. Supports bond, pocket, and contact entries using the chain IDs above.

Templates (JSON, optional) Native Boltz templates array. Each entry may select a CIF/PDB, chain mapping, force flag, and threshold.

Properties (JSON, optional) Native Boltz properties array, such as [{"affinity": {"binder": "B"}}].

Boltz input (YAML) Complete native Boltz YAML. Define protein, DNA, RNA, and ligand entities under sequences; optional constraints, templates, and affinity properties use the same schema as the Boltz CLI. A protein without an msa key gets one from the MSA server; give msa a path to supply your own, or msa: empty for single-sequence mode. In this form: Bio Web writes this text to input.yaml and passes that file to boltz predict.

Boltz input file (YAML) The YAML file defining this prediction, in the same schema as the Boltz CLI takes. In this form: Choose a file. Its contents are read here and sent as the input document.

Inference

Use the public MSA server Generate an MSA for every protein that has no MSA of its own, using the ColabFold MMseqs2 server below. This sends those protein sequences to that server. Proteins with an MSA path or msa: empty are left as they are. Turn off for private sequences or offline runs; each protein then needs a path or msa: empty.

MSA server URL ColabFold-compatible MMseqs2 endpoint; used only when the MSA server is on. The public server asks for serial queries from a single IP, so host your own for large batches.

Use inference-time potentials Apply Boltz-2 inference-time potentials to improve physical plausibility.

Recycling steps Number of recycling steps. The Boltz CLI default is 3.

Structure sampling steps Diffusion steps used for structure prediction. The Boltz CLI default is 200.

Structure samples Independent diffusion samples generated for each input.

Structure step scale Diffusion temperature/step size. Boltz-2 defaults to 1.5 and recommends values between 1 and 2; lower values increase diversity.

Affinity

Affinity sampling steps Sampling steps for the affinity head. Used only when the YAML requests an affinity property.

Affinity samples Diffusion samples for the affinity head. The Boltz CLI default is 5.

Apply affinity molecular-weight correction Enable Boltz's molecular-weight correction for the affinity value head.

Output

Random seed Seed used by the Boltz random-number generator.

Coordinate format Format of predicted coordinate files.

Ready