Boltz (Deprecated)¶
Runs legacy Boltz structure prediction on protein sequences using MSA inputs. Generates multiple sampled structures and corresponding confidence/quality metrics. This node is deprecated; use Boltz-2 nodes for enhanced features.
Usage¶
Use this node only for legacy workflows that still depend on the original Boltz service. Provide MSA search results (A3M) for one or more sequences; the node will perform structure prediction with configurable recycling and diffusion sampling, returning ranked PDBs and confidence scores. For new projects, migrate to Boltz2StructurePredictionNode.
Inputs¶
| Field | Required | Type | Description | Example |
|---|---|---|---|---|
| a3m | True | A3M | MSA search results as a dictionary mapping sequence IDs to A3M-formatted strings. Each entry will be processed to predict structures. | {'seq1': '>seq1\nAAAA...\n>seq1/2-100\n-AAA...', 'seq2': '>seq2\nMKTW...\n>seq2/align\n-MKT...'} |
| recycling_steps | True | INT | Number of recycling iterations during prediction. Higher values can improve accuracy but increase runtime. | 10 |
| diffusion_samples | True | INT | Number of structure samples to generate via diffusion for each input sequence. More samples yield more diverse predictions. | 5 |
| seed | True | INT | Base random seed. If multiple sequences are processed, each uses an incremented seed (seed + index). | 42 |
| mode | True | STRING | Execution mode: MOCK (uses predefined mock outputs), PROD (runs the live service), TEST (overrides key params for quick checks). | PROD |
Outputs¶
| Field | Type | Description | Example |
|---|---|---|---|
| folding.pdb | PDB | Ranked predicted structures as a dictionary mapping ' | {'seq1_rank_001': 'MODEL 1\nATOM ...\nENDMDL\n', 'seq1_rank_002': 'MODEL 1\nATOM ...\nENDMDL\n'} |
| confidence_scores.json | JSON | Confidence and related quality metrics for each predicted structure, keyed by ' | {'seq1_rank_001': {'pLDDT_mean': 78.4, 'PAE': [[0.0, 1.2], [1.1, 0.0]]}, 'seq1_rank_002': {'pLDDT_mean': 74.2, 'PAE': [[0.0, 2.0], [2.1, 0.0]]}} |
Important Notes¶
- Deprecated: This node is deprecated. Prefer Boltz2StructurePredictionNode for YAML input support, ligands, DNA/RNA, potentials, and affinity prediction.
- Mode behavior: TEST mode forces recycling_steps=1 and diffusion_samples=1 to reduce runtime. MOCK mode returns predefined mock results.
- Multiple inputs: When multiple sequences are provided, the node processes each and increments the seed per sequence.
- Performance: Increasing recycling_steps and diffusion_samples increases runtime and cost.
- Input expectations: a3m must be a dict of A3M-formatted strings keyed by sequence IDs; each must contain a valid FASTA header and aligned sequences.
Troubleshooting¶
- Empty or missing outputs: Ensure 'a3m' is a non-empty dict and each value is valid A3M text with a FASTA header line.
- Service timeout or long runtime: Reduce diffusion_samples and recycling_steps, or test with mode='TEST' to validate the pipeline quickly.
- Inconsistent IDs in results: Output keys are derived from input sequence IDs; verify your 'a3m' dict keys are unique and meaningful.
- Unexpectedly few structures: Check diffusion_samples; in TEST mode it is forced to 1.
- Reproducibility issues: Confirm the base seed and remember seeds are incremented per sequence (seed + index).
- Mock data confusion: If using mode='MOCK', outputs are fixed from mock data and will not reflect your inputs.