STRING (Protein-Protein Interaction Networks)

STRING is a biological database and web resource of known and predicted protein-protein interactions — functional associations between proteins, not only direct physical binding but also indirect relationships such as participating in the same pathway or complex. It is maintained by an academic consortium spanning the Novo Nordisk Foundation Center for Protein Research, EMBL, the University of Copenhagen, the Swiss Institute of Bioinformatics, Dresden Technical University, and the University of Zurich, and covers a genuinely large span of biology: version 12.0 (2023) reports data on roughly 59 million proteins across more than 12,000 organisms.

How STRING scores an interaction

What makes STRING distinct from a single-source interaction list is that every edge in the network carries a weighted confidence score, assembled from several independent evidence channels rather than one method alone: interactions imported from curated experimental databases (BioGRID, HPRD, MINT among others), automated text mining across the literature (PubMed abstracts, model-organism databases like SGD and FlyBase) for statistically significant gene co-occurrence, and purely computational predictions drawn from genomic context — conserved gene neighborhood, gene fusion events across genomes, phylogenetic co-occurrence patterns — and expression coexpression data. In the interactive network viewer, edge color saturation directly represents that combined confidence score, so a researcher can visually distinguish a well-supported, multi-evidence interaction from one that rests on text-mining co-occurrence alone. All predictions are benchmarked against a common reference of curated functional partnerships, including KEGG pathway membership, to calibrate what a given confidence level actually means.

Why the evidence-channel distinction matters

Because STRING deliberately combines heterogeneous evidence types into one network, the confidence score is the mechanism that keeps a text-mined co-occurrence from being presented with the same weight as a directly observed physical interaction. A researcher building a candidate interactome for downstream experimental validation should filter to a defined confidence threshold and, ideally, note which evidence channels contributed to the interactions retained — treating every STRING edge as equivalent evidence is a real methodological error the scoring system is specifically built to let a user avoid, not one it prevents automatically regardless of how the data is used.

Access

STRING is freely accessible through its website for interactive network exploration, as bulk downloadable files for full offline analysis, and through a programmatic API for integration into automated pipelines. No account or paid tier gates access to the core interaction data, unlike some other named bioinformatics resources that have introduced subscription requirements for bulk use.

Where STRING fits among named repositories

STRING is a derived, multi-source association network, not a primary experimental repository — it draws on and cross-references curated pathway resources like KEGG and Reactome as part of its own evidence integration, rather than existing independently of them. A methods section or data management plan describing a STRING-based analysis should specify the confidence-score threshold applied and, where the distinction matters to the claim being made, which evidence channels were included — the same accession- and parameter-level specificity CASRAI’s own Data Management Plan guidance recommends for any named repository dependency.