MSA Search¶
Runs multiple sequence alignment (MSA) searches for one or more protein sequences. It queries managed databases, merges results from multiple sources, and returns per-sequence alignments in A3M format. Supports MOCK, PROD, and TEST modes, including automatic format conversion and result consolidation.
Usage¶
Use this node at the start of a protein structure prediction workflow to generate MSAs from input FASTA sequences. Select a database preset based on speed vs. coverage needs (toy, reduced, or full). Feed the resulting A3M output into downstream folding/prediction nodes.
Inputs¶
| Field | Required | Type | Description | Example |
|---|---|---|---|---|
| fasta | True | FASTA | One or more protein sequences in FASTA format. Each sequence will be processed independently and returned as a separate A3M entry. | >seq1 MKTAYIAKQRQISFVKSHFSRQDILDLWQ >seq2 GHHHHHHSSGVDLGTENLYFQSMASMTGGQQ |
| db_preset | True | STRING (enum) | Which database set to search. Allowed values: alphafold_toy_dataset (fast, minimal), alphafold_reduced_dataset (balanced), alphafold_full_dataset (maximum coverage). | alphafold_reduced_dataset |
| mode | True | STRING (enum) | Execution mode. MOCK uses bundled mock data, PROD calls the live service, TEST forces minimal settings for quick checks. | PROD |
Outputs¶
| Field | Type | Description | Example |
|---|---|---|---|
| msa.a3m | MSA | Dictionary mapping each input sequence ID to its merged A3M alignment text. | {"seq1": "# A3M content...\n>seq1\nMKTAYI-...", "seq2": ">seq2\nG-HHHS..."} |
Important Notes¶
- In TEST mode, the database preset is automatically overridden to alphafold_toy_dataset for faster turnaround.
- Timeouts scale with the number of input sequences and the selected database preset (toy < reduced < full). Very large batches can take a long time.
- Results from multiple sources are merged: Stockholm-formatted MSAs are aggregated and converted to A3M per sequence.
- If a source returns A3M, it is internally converted to Stockholm before merging to ensure consistent processing.
- MOCK mode serves precomputed responses from bundled mock data and does not perform real searches.
- Ensure unique FASTA headers; output keys mirror the input sequence IDs.
- The node returns an A3M dictionary even if some sequences have no hits; such entries may be empty.
Troubleshooting¶
- Empty or missing A3M for a sequence: Verify the sequence ID is unique and valid. Try a larger database preset (reduced or full) for better coverage.
- Timeouts or job takes too long: Reduce batch size (fewer sequences per run), use alphafold_toy_dataset or alphafold_reduced_dataset, or run sequences in separate executions.
- Unexpected MOCK-like results: Confirm mode is set to PROD (not MOCK or TEST).
- Downstream node rejects output: Ensure you're passing the MSA (A3M) dictionary directly; do not modify keys or formats.
- Mismatched sequence IDs: Check FASTA headers for duplicates or unusual characters; headers become dictionary keys in the output.