Protein Structure Pipeline
Overview
Chains VenusFactory structure tools: AlphaFold DB download + confidence analytics, local ESMFold when no UniProt ID, RCSB for experimental structures, and PyMOL rendering.
VenusFactory execution
Call hub tools only. Large PDB/mmCIF stay on disk via file_info.file_path.
Project Tools (VenusFactory2)
| Tool | Args | When |
|---|---|---|
| download_alphafold_structure_by_uniprot_id | uniprot_id, out_dir, format |
Known UniProt accession |
| download_alphafold_metadata_by_uniprot_id | uniprot_id, out_dir |
Need pLDDT metadata JSON |
| analyze_alphafold_plddt_by_metadata_file | metadata_path |
Confidence per residue |
| analyze_alphafold_pae_by_pae_file | pae_path |
Domain/interface confidence |
| predict_structure_esmfold | sequence, optional output_dir |
No UniProt / quick local fold |
| download_rcsb_structure_by_pdb_id | pdb_id, out_dir, format |
Experimental structure |
| download_rcsb_entry_metadata_by_pdb_id | pdb_id, out_path |
Resolution, method, ligands |
| render_protein_structure | pdb_path, out_dir, style options |
Publication-quality still |
| superpose_two_structures | pdb_a, pdb_b, out_dir |
Compare models |
Workflow
UniProt → AlphaFold (default)
download_alphafold_structure_by_uniprot_iddownload_alphafold_metadata_by_uniprot_id→analyze_alphafold_plddt_by_metadata_file- If PAE available,
analyze_alphafold_pae_by_pae_file - Optional
render_protein_structure(cartoon + pLDDT coloring if supported by tool args)
Sequence-only → ESMFold
- Obtain sequence (
read_fasta/ UniProt seq tool). predict_structure_esmfold- Optional RSA/SASA/SS via
protein_property_predictiontools.
Experimental PDB
download_rcsb_structure_by_pdb_id+ metadata.- Prefer over AlphaFold when an experimental entry exists for the same construct.
When NOT to use
- Structural homolog search with active-site masking →
foldseek_structural_similarity - Domain annotation without structure →
interpro_domain_annotation
Common mistakes
- Using AlphaFold when user already has a high-res PDB.
- Dumping PDB text into chat instead of using
file_info.file_path. - Skipping pLDDT before trusting loop regions for mutation design.