The STRING database in 2025: protein networks with directionality of regulation.
Szklarczyk D, Nastou K, Koutrouli M, Kirsch R, Mehryary F, Hachilif R, Hu D, Peluso ME, Huang Q, Fang T, Doncheva NT, Pyysalo S, Bork P, Jensen LJ, von Mering C. · 2025
Read The STRING database in 2025: protein networks with directionality of regulation. by Szklarczyk D, Nastou K, Koutrouli M, Kirsch R, Mehryary F, Hachilif R, Hu D, Peluso ME, Huang Q, Fang T, Doncheva NT, Pyysalo S, Bork P, Jensen LJ, von Mering C. (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/europmc-pmc11701646.txt.
Introduction
Szklarczyk and colleagues update the STRING database of protein networks in this 2025 paper. Proteins work by binding, regulating, and cooperating with each other. STRING compiles, scores, and merges evidence on protein-protein links from experiments, predictions, and prior knowledge. Its goal is a full, objective global network of physical and functional links.
Version 12.5 adds a new regulatory network. For it, the team gathers evidence on the type and direction of each regulatory link, using curated pathway databases, plus a language model fine-tuned to parse the literature. Users can now view three network types on their own: functional, physical, and regulatory. The pathway enrichment tool, which flags processes over-represented in a protein set, was also updated. Network clustering remains part of the package.
Catalog record
| CREATOR | Szklarczyk D, Nastou K, Koutrouli M, Kirsch R, Mehryary F, Hachilif R, Hu D, Peluso ME, Huang Q, Fang T, Doncheva NT, Pyysalo S, Bork P, Jensen LJ, von Mering C. |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 45 KB |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
The STRING database in 2025: protein networks with directionality of regulation By Damian Szklarczyk, Katerina Nastou, Mikaela Koutrouli, Rebecca Kirsch, Farrokh Mehryary, Radja Hachilif, Dewei Hu, Matteo E Peluso, Qingyao Huang, Tao Fang, Nadezhda T Doncheva, Sampo Pyysalo Published in: Nucleic Acids Research DOI: 10.1093/nar/gkae1113 ABSTRACT Abstract Proteins cooperate, regulate and bind each other to achieve their functions. Understanding the complex network of their interactions is essential for a systems-level description of cellular processes. The STRING database compiles, scores and integrates protein–protein association information drawn from experimental assays, computational predictions and prior knowledge. Its goal is to create comprehensive and objective global networks that encompass both physical and functional interactions. Additionally, STRING provides supplementary tools such as network clustering and pathway enrichment analysis. The latest version, STRING 12.5, introduces a new ‘regulatory network’, for which it gathers evidence on the type and directionality of interactions using curated pathway databases and a fine-tuned language model parsing the literature. This update enables users to visualize and access three distinct network types—functional, physical and regulatory—separately, each applicable to distinct research needs. In addition, the pathway enrichment detection functionality has been updated, with better false discovery rate corrections, redundancy filtering and improved visual displays. The resource now also offers improved annotations of clustered networks and provides users with downloadable network embeddings, which facilitate the use of STRING networks in machine learning and allow cross-species transfer of protein information. The STRING database is available online at https://string-db.org/. INTRODUCTION The function of every living cell is primarily governed by a complex network of interacting proteins, with each protein’s role determined not only by its molecular activities but also by its position within this network (1,2). Connected proteins work together to contribute to common biological processes through various interaction types, such as physical binding, genetic interactions and regulatory influences. These interactions can collectively be categorized as functional associations, which serve as fundamental operational units within biological systems. Unraveling protein networks in their various modalities remains a significant research focus. Consequently, numerous databases have been developed over the years to meet specific research needs. These range from carefully curated pathway databases such as Reactome (3) and KEGG (4), to databases of experimental interaction evidence curated from literature such as BioGRID (5), IntAct (6) and MINT (7), and to those centered on specific modes of interactions, including Complex Portal for protein co-complexes (8) and SIGNOR for regulatory interactions (9). Finally, composite databases such as STRING (10), GeneMANIA (11), FunCoup (12) and HumanNet (13) not only incorporate data from these sources but also employ an array of computational methods to predict additional associations, striving to provide the most comprehensive views of the interactome. Among these databases, STRING is notable for its many sources of evidence, its robust scoring system, user-friendly interface and comprehensive suite of enrichment features. It is dedicated to assembling a broad set of associations among proteins for the complete proteomes of thousands of organisms across all domains of life. STRING charts interactions ranging from highly confident, well-documented associations to more speculative ones, which are crucial for exploratory and computational analyses. The scoring system ensures that data from diverse sources—including automatic text mining, high- and low-throughput experimental data, and computational predictions—are directly comparable and weighted consistently, regardless of their origin. Additionally, STRING extends and maps networks across species by predicting interologs, thereby broadening the scope of functional associations to encompass a diverse range of organisms. This includes uncultured novel bacterial species derived from metagenomic samples, and even unpublished proteomes uploaded by users. The development of interaction databases has so far been a trade-off between detailing the exact mode of an interaction—whether physical or functional, signaling or structural, stable or transient—and maintaining a comprehensive set of interactions. Given the limitations of the available data and methodologies, STRING has historically focused on broadly defined functional associations, which provided the most useful set of interactions for proteome-wide analysis. While these networks continue to perform well in downstream tasks (14,15), they often lack fine-grained resolution, leaving more specialized databases better equipped to offer detailed insights into specific interactions if needed. However, the growing volume of data and advancements in text-mining technologies have since enabled the composite, general-purpose databases to catch up, offering more detailed maps of the interaction space. Recently, the STRING database introduced a co-complex (physical) interaction network mode, which details multi-protein assemblies. With the latest update, the database now also includes the largest set of regulatory (directed) interactions, highlighting the flow of information within cells. This major enhancement greatly expands the analytical capabilities of the database, enabling new types of studies that were not previously possible with nondirected interaction networks. Interaction and pathway databases are frequently used to gain insights into the functional context of individual proteins or to understand the biological organization of an entire protein dataset. Many of these databases enhance the interpretation of protein datasets beyond simple interaction annotations by incorporating an enrichment analysis toolset. This type of analysis involves comparing observed data against expected distributions, enabling researchers to identify statistically significant features or patterns. Enrichment tools typically utilize established ontologies and annotations, such as Gene Ontology (16), MSigDB hallmark sets (17) or OMIM (18). A few databases also leverage their annotated datasets to help understand the functional context of a user’s input. For example, KEGG (4) employs pathway and module abstractions to organize genes into datasets for overrepresentation analysis, while Reactome maps interconnected proteins into hierarchically nested modules. Uniquely, STRING uses an unsupervised procedure to hierarchically precluster its entire proteome-wide networks into functional modules. Incorporating these network-derived gene sets into the enrichment analysis facilitates the identification of novel modules, especially in areas of the proteome where manual curation has been less comprehensive so far. DATABASE CONTENT The basic interaction scope in STRING is that of a ‘functional association’ between pairs of proteins. A functional association is defined as a contribution of two non-identical proteins to a common function (19,20). This can take many forms; functionally associated proteins can be in physical proximity to each other, regulate each other, exhibit genetic epistasis or even work antagonistically (as long as this occurs in the context of a common function). For the purpose of defining functional associations, the concept of a common function is crucial, but difficult to define—it should broadly be thought of as corresponding to the notion of a ‘pathway’ or ‘function module’. In practice, STRING roughly follows the functional granularity of the pathway maps in the KEGG database (4). From the set of all functional associations in STRING, subsets are derived that are more specifically annotated with regard to their mechanism of association. Currently, two such more specific subsets are implemented: ‘physical’ and ‘regulatory’. The physical mode refers to pairs of proteins that either bind directly or are at least subunits of the same complex (21). The regulatory mode refers to associated protein pairs that are known to regulate each other’s activity in at least one direction. This mode is described in more detail further below; for the first time in STRING, such connections are annotated not only with confidence scores but also with a directionality. All protein–protein associations in STRING are annotated with ‘confidence scores’. These scores are fully precomputed; they scale between 0 and