STRING Protein Interaction Network

Tier: 1 — Essential | Category: Protein-Protein Interaction Database

STRING is the world’s most comprehensive database of known and predicted protein-protein associations, covering >14,000 organisms and integrating seven evidence channels: co-expression, genomic context, co-occurrence, experimentally determined interactions, curated databases, text mining, and protein homology ^{1The STRING database in 2023: protein-protein association networks and functional enrichment analyses for any sequenced genome of interest2023 · Nucleic Acids Research · DOI 10.1093/nar/gkac1000 · PMID 36370105Open reference}. The STRING v12.0 database covers >67 million proteins and enables large-scale network analyses for any sequenced genome.

Evidence Channels and Confidence Scores

STRING assigns a confidence score (0–1000) to each protein association across seven channels:

Channel	Source
Neighborhood	Genomic co-localization
Co-occurrence	Phylogenetic co-occurrence
Co-expression	mRNA co-expression across datasets
Experiments	Yeast 2-hybrid, pull-downs, AP-MS
Databases	KEGG, Reactome, IntAct, MINT
Text mining	Co-occurrence in PubMed abstracts
Homology	Conserved interactions in model organisms

Only associations with combined score ≥ 400 (medium confidence) are typically used for hub analysis; ≥ 700 (high confidence) for mechanistic claims.

Applications in Neurodegeneration Research

STRING network analysis has become a standard step in neurodegeneration research for:

Hub gene identification: Identifying highly connected nodes in disease-associated gene networks to prioritize therapeutic targets. For Alzheimer’s disease, STRING analysis consistently identifies APP, PSEN1, MAPT, APOE, TREM2, and TP53 as network hubs ^{2A Meta-analysis to Identify Common Key Genes Across Ageing, Alzheimer's Disease and other Neurodegenerative Disorders2026 · Annals of Neurosciences · DOI 10.1177/09727531261420619Open reference}.

Proteomics integration: Mapping mass-spectrometry-identified proteins onto the STRING interactome to identify sub-network modules perturbed in disease, as demonstrated in the proteogenomic analysis of Parkinson’s disease postmortem brain ^{3Mitochondria dysfunction in the pathogenesis of Alzheimer's disease: recent advances2020 · Molecular Neurodegeneration · DOI 10.1186/s13024-020-00376-6Open reference}.

Drug target network analysis: Computing betweenness centrality and PageRank in STRING-derived networks to identify proteins whose inhibition would maximally disrupt disease-associated signaling modules.

Multiprotein complex visualization: The STRING Cytoscape plugin (stringApp) enables integration of STRING networks with transcriptomics and proteomics data in Cytoscape for co-visualization ^{4Cytoscape stringApp 2.0: Analysis and Visualization of Heterogeneous Biological Networks2022 · Journal of Proteome Research · DOI 10.1021/acs.jproteome.2c00651Open reference}.

Tutorial: STRING network analysis for TREM2 interactome

import requests
import json

# STRING API v12.0
STRING_API = "https://version-12-0.string-db.org/api"

# Get TREM2 interaction partners
params = {
    "identifiers": "TREM2",
    "species": 9606,  # Homo sapiens
    "caller_identity": "scidex_forge",
    "required_score": 700,  # high confidence
    "limit": 20,
}
resp = requests.get(f"{STRING_API}/json/interaction_partners", params=params)
interactions = resp.json()

# Extract interaction partners
partners = [(i["preferredName_B"], i["score"]) for i in interactions]
partners_sorted = sorted(partners, key=lambda x: -x[1])

print("Top TREM2 interaction partners (STRING score ≥ 700):")
for name, score in partners_sorted[:10]:
    print(f"  {name}: {score:.0f}")

# Output typically includes: DAP12/TYROBP, APOE, PLCG2, PIK3CA, SYK, BTK, VAV1

Neurodegeneration-Relevant PPI Networks in STRING

The SciDEX string-protein-interactions skill wraps the STRING API to retrieve interaction partners and enrichment results. Key network modules identified by STRING in neurodegeneration include:

• TREM2-DAP12-PLCG2 microglial signaling hub (Alzheimer’s)

• PINK1-Parkin-VDAC1 mitophagy complex (Parkinson’s)

• TDP-43-FUS-hnRNP RNA-binding protein network (ALS/FTD)

• APP-PSEN1-PSEN2-APOE amyloid processing complex (Alzheimer’s)

Cross-reference: analysis [[8ec36980-febb-4093-a5a1-387ea5768480]] (Proteogenomic Network Hubs as Druggable Targets in Early PD Neurodegeneration) used STRING to identify hub proteins for therapeutic prioritization.

Relevance to SciDEX

The STRING network underpins the knowledge graph’s protein-protein interaction edges. The string-protein-interactions Forge skill enables on-demand retrieval of interactors for any protein of interest, with confidence scores propagated as KG edge weights. Exchange hypotheses targeting protein-protein interactions can use STRING scores as prior evidence for the existence and tractability of the proposed interaction.

Limitations

• Text-mining channel can introduce false positives when proteins are frequently co-mentioned without direct interaction

• Interaction scores reflect association probability, not interaction directionality or mechanism

• Many interactions are context-dependent (tissue, cell state, condition) but STRING does not capture context

• Coverage drops substantially for less-studied proteins; neurodegeneration-relevant hits from GWAS may have sparse STRING coverage

• STRING does not distinguish activating from inhibitory interactions

Cross-References

• [[ai-tool-reactome]] — Pathway context for STRING-identified modules

• [[mechanisms-convergent-pathways-neurodegeneration]] — Pathways for which STRING defines the core PPI structure

• [[ai-tool-alphafold3]] — Structural validation of STRING-predicted complexes