Source: https://github.com/aipoch/medical-research-skills

AlphaFold Database Skill

When to Use

• You have a UniProt accession (e.g., P00520) and need to download its AlphaFold-predicted 3D structure in mmCIF or PDB format.
• You want to assess prediction reliability using per-residue pLDDT confidence scores.
• You need PAE (Predicted Aligned Error) data to support domain-level interpretation or inter-domain confidence checks.
• You want to programmatically retrieve AlphaFold DB metadata (e.g., model/structure URLs and related fields) for downstream pipelines.
• You are building an automated workflow that fetches structures + confidence metrics for many proteins by UniProt ID.

Key Features

• Fetch AlphaFold DB predicted structures by UniProt accession.
• Download structure files in mmCIF (default) or PDB.
• Retrieve and save metadata JSON, including confidence-related fields (e.g., pLDDT) and URL information.
• Simple CLI workflow suitable for scripting and batch processing.

Dependencies

• Python >=3.8
• requests >=2.25

Example Usage

Fetch the structure and metadata for a UniProt ID and save them to a directory:

python scripts/fetch_structure.py --uniprot_id P00520 --output_dir ./out --format cif

Fetch as PDB instead:

python scripts/fetch_structure.py --uniprot_id P00520 --output_dir ./out --format pdb

Expected outputs in ./out:

• P00520.cif (or P00520.pdb)
• P00520_metadata.json (includes confidence/URL fields such as pLDDT-related information)

Implementation Details

• Input identifier: UniProt accession ID (e.g., P00520).
• Formats:
  • --format cif (default): downloads an mmCIF structure file.
  • --format pdb: downloads a PDB structure file.
• Artifacts written:
  • Structure file named <UNIPROT_ID>.<cif|pdb>.
  • Metadata JSON named <UNIPROT_ID>_metadata.json, used to store confidence metrics (e.g., pLDDT) and AlphaFold DB URL-related fields.
• API reference: Endpoint and response details are documented in references/api_reference.md.