medsci-agent: Skill for OpenCode

AlphaFold Database

Overview

AlphaFold DB (by DeepMind and EMBL-EBI) provides predicted 3D structures for over 200M proteins. Structures are predicted by AlphaFold2 and stored with per-residue confidence scores (pLDDT).

API Access

import requests

# Fetch prediction for a UniProt accession
uniprot_id = "P04637"
r = requests.get(f"https://alphafold.ebi.ac.uk/api/prediction/{uniprot_id}")
prediction = r.json()[0]

# Get model URLs
cif_url = prediction["cifUrl"]        # mmCIF format
pdb_url = prediction["pdbUrl"]        # PDB format
pae_url = prediction["paeImageUrl"]   # PAE plot image

Download Structure

# Direct download
pdb_url = f"https://alphafold.ebi.ac.uk/files/AF-{uniprot_id}-F1-model_v4.pdb"
cif_url = f"https://alphafold.ebi.ac.uk/files/AF-{uniprot_id}-F1-model_v4.cif"
pae_url = f"https://alphafold.ebi.ac.uk/files/AF-{uniprot_id}-F1-predicted_aligned_error_v4.json"

pLDDT Confidence Score

• Stored in the B-factor column of PDB files.
• >90: High confidence (blue). Reliable backbone and side-chain.
• 70-90: Confident (cyan). Good backbone prediction.
• 50-70: Low confidence (yellow). Caution with interpretation.
• <50: Very low (orange). Likely disordered or uncertain.

Predicted Aligned Error (PAE)

• Matrix of expected position error between all residue pairs.
• Low PAE between domains indicates confident relative orientation.
• High PAE between domains means they may be flexible or uncertain.

Key Details

• One model per UniProt accession (longest isoform, up to 2700 residues).
• Structures lack ligands, cofactors, and post-translational modifications.
• Use pLDDT to filter reliable regions before docking or analysis.
• For custom sequences not in the DB, run AlphaFold2 or use ESMFold.