Boltz Multimer¶
Predicts binder–target multimer protein complexes using Boltz-based diffusion with recycling. Takes MSA inputs for a binder and a single target, optionally focuses on specified target interface residues, and returns ranked complex PDBs with confidence metrics. Deprecated: prefer Boltz2MultimerNode for newer features and continued support.
Usage¶
Use this node after generating MSAs for both a binder and a single target sequence. Provide the binder's A3M(s) and exactly one target A3M. Optionally specify a set of target residue indices to guide interface contacts. The node will run multimer prediction and return ranked complex structures and confidence scores for each binder entry.
Inputs¶
| Field | Required | Type | Description | Example |
|---|---|---|---|---|
| binder_a3m | True | A3M | Dictionary of binder MSA(s) in A3M format. Keys are sequence IDs; values are A3M contents. Each binder entry will be paired against the single target. | {"binder_1": " |
| target_a3m | True | A3M | Dictionary containing exactly one target MSA in A3M format. The single key is the target sequence ID; the value is its A3M content. | {"target": " |
| recycling_steps | True | INT | Number of recycling iterations during prediction. Higher values may improve accuracy at the cost of runtime. | 10 |
| diffusion_samples | True | INT | Number of diffusion samples (structure hypotheses) to generate per binder entry. Larger values yield more diversity. | 5 |
| target_specs | True | STRING | Comma-separated list of residue indices on the target chain that define or focus the binding interface. | 1,2,3,45,78 |
| seed | True | INT | Base random seed. Each binder entry increments this seed to ensure varied yet reproducible sampling. | 42 |
Outputs¶
| Field | Type | Description | Example |
|---|---|---|---|
| folding.pdb | PDB | Dictionary of ranked predicted complex structures. Keys combine binder ID and rank; values are PDB contents. | {"binder_1_rank_1.pdb": " |
| confidence_scores.json | JSON | Dictionary of confidence metrics per predicted structure (e.g., model confidence and related scores). | {"binder_1_rank_1": {"pLDDT": 85.2, "other_metric": 0.91}} |
Important Notes¶
- Deprecated: This node is deprecated. Use Boltz2MultimerNode for improved functionality (constraints, templates, multiple entities, affinity prediction).
- Single target only: target_a3m must contain exactly one entry; multiple targets are not supported.
- Interface residues: target_specs must be integers (1-based indices) separated by commas; invalid or empty values may degrade or invalidate results.
- Batch behavior: All binder entries are paired with the single target; the seed is incremented per binder entry.
- Runtime considerations: Higher recycling_steps and diffusion_samples increase compute time.
Troubleshooting¶
- Multiple target entries provided: If you pass more than one target A3M, the node raises an error. Provide exactly one key-value pair in target_a3m.
- Invalid target_specs: If residue indices cannot be parsed as integers, correct the input to a comma-separated list of integers (e.g., 5,12,27).
- Mismatched or malformed A3M data: Ensure binder_a3m and target_a3m contain valid A3M text for the intended sequences; regenerate MSAs if necessary.
- Empty outputs or missing ranks: Reduce diffusion_samples or recycling_steps to test minimal settings, verify input MSAs are not empty, and retry.
- Timeouts or long runtimes: Lower recycling_steps and diffusion_samples; run fewer binder entries per invocation.