finding-protein-data

SOP: finding protein and gene data online

For any protein/gene page, pull from these primary sources. Cross-reference UniProt + NCBI Gene at minimum.

Primary databases

UniProt — https://www.uniprot.org/

Best for: Curated protein-level information — sequence, function, domains, post-translational modifications, disease associations, all with PubMed-cited evidence.

What to extract:

• UniProt accession (e.g., P04637 for human p53)
• Recommended name + alternative names
• Functional summary (the “Function” section)
• Subcellular location
• Domain structure
• Key PTMs (phosphorylation sites, etc.)
• Disease/phenotype associations

Reviewed vs unreviewed: Use Swiss-Prot (manually curated) entries — accession starts with a letter and is short. TrEMBL entries (auto-generated) are less reliable.

Citation: UniProt P04637 (TP53_HUMAN), accessed YYYY-MM-DD

NCBI Gene — https://www.ncbi.nlm.nih.gov/gene/

Best for: Gene-level info — genomic location, exon structure, RNA isoforms, expression data (linked to GTEx), orthologs across species.

What to extract:

• NCBI Gene ID (e.g., 7157 for human TP53)
• HGNC symbol (TP53)
• Official full name
• Genomic location and size
• Orthologs (especially mouse, useful for cross-species claims)
• RefSeq transcripts

Citation: NCBI Gene ID 7157, accessed YYYY-MM-DD

HGNC — https://www.genenames.org/

Best for: Authoritative human gene nomenclature. Use HGNC symbols in filenames and frontmatter to avoid synonym ambiguity.

Interaction databases

STRING — https://string-db.org/

Best for: Protein-protein interaction network with confidence scoring (text-mining, experimental, database, co-expression evidence channels).

What to extract:

• Top 10–20 interactors at confidence ≥0.7
• Distinguish “experimental” from “predicted” (text-mining-only) edges

Caveat: STRING’s text-mining channel is noisy. For mechanistic claims, require either “experimental” or “database” evidence.

BioGRID — https://thebiogrid.org/

Best for: Manually curated physical and genetic interactions, with the original publication for each.

What to extract: Direct interactors with PubMed IDs for each interaction → ingest the most-cited interaction studies into studies/.

IntAct — https://www.ebi.ac.uk/intact/

Best for: Detailed interaction evidence with experimental method (Y2H, AP-MS, IP, etc.). More structured than STRING for evaluating interaction quality.

Function and ontology

AmiGO / Gene Ontology — https://amigo.geneontology.org/

Best for: Standardized functional annotations (molecular function, biological process, cellular component). Use for “what does this protein do” at a controlled-vocabulary level.

InterPro — https://www.ebi.ac.uk/interpro/

Best for: Protein domains and motifs.

CAZy — https://www.cazy.org/

Best for: Carbohydrate-active enzyme (CAZyme) families and modules, especially when evaluating glycoside hydrolases (GHs), glycosyltransferases (GTs), polysaccharide lyases (PLs), carbohydrate esterases (CEs), auxiliary activities (AAs), and carbohydrate-binding modules (CBMs). Use for enzyme-family scouting when a hypothesis involves carbohydrate or glycoconjugate recognition/catalysis, such as engineered glucosepane-cleavage scaffolds.

What to extract:

• CAZy family or module class (e.g., GH, GT, PL, CE, AA, CBM)
• Family number and clan where available
• Known activities and EC numbers
• Mechanism annotation if listed
• Organism distribution and candidate microbial/fungal homologs
• Cross-references to UniProt, GenBank, PDB, EC, and PubMed

Caveat: CAZy family membership supports fold/function hypothesis generation, not proof that a specific enzyme cleaves a proposed substrate. For noncanonical substrates such as glucosepane, require direct biochemical validation on authentic substrate; do not infer activity from GH/CBM family membership alone.

Citation: CAZy family GHxx / CBMxx, accessed YYYY-MM-DD

Workflow for a new protein page

1. UniProt → accession, function summary, domains, PTMs, disease associations.
2. NCBI Gene → gene ID, HGNC symbol, mouse/rat orthologs.
3. InterPro / GO → domains and controlled-vocabulary function.
4. CAZy → enzyme-family/module annotations when the protein is a CAZyme or candidate carbohydrate/glycoconjugate-active enzyme.
5. STRING (confidence ≥0.7, experimental evidence channel) → top interactors.
6. List pathways the protein participates in → wikilink to pathway pages (run sops/finding-pathway-data.md if pathway pages don’t exist yet).
7. Map to relevant Hallmarks of Aging.
8. Use archive search to find the 3–5 most-cited reviews of the protein’s role in aging → create study pages.

Frontmatter example

---
type: protein
aliases: [TP53, tumor protein p53, "guardian of the genome"]
uniprot: P04637
ncbi-gene: 7157
hgnc: 11998
mouse-ortholog: Trp53
ensembl: ENSG00000141510
key-domains: [DNA-binding, tetramerization, transactivation]
key-ptms: [Ser15-phos, Ser20-phos, Lys382-acetyl]
pathways: ["[[p53-pathway]]", "[[dna-damage-response]]", "[[apoptosis-pathway]]"]
hallmarks: ["[[genomic-instability]]", "[[cellular-senescence]]"]
known-interactors: ["[[mdm2]]", "[[cbp-p300]]", "[[atm]]"]
---

What NOT to trust

• Wikipedia protein boxes (good starting point only).
• “Predicted” annotations not backed by experimental evidence.
• STRING text-mining-only edges for mechanistic claims.
• Old NCBI Gene records — check the “Update date” at the bottom.

See also

• finding-pathway-data — for pathway-level context
• finding-compound-data — for inhibitors/agonists of the protein