Protein Interactions · LiteFold
STRING
A database of physical and functional protein-protein interactions with sequence and organism metadata.
What STRING is
This dataset presents a structured distribution of STRING version 12.0, prepared by LiteFold. STRING is an established resource mapping physical and functional associations between proteins across a vast range of organisms. It collates direct physical bindings as well as indirect metabolic and regulatory associations to help researchers understand cellular pathways and systems biology.
Deriving comprehensive interaction maps typically requires harmonising disparate data types. STRING addresses this by aggregating laboratory experiments, computational prediction algorithms, text-mined scientific literature, and manually curated knowledge bases. Each interaction is evaluated across these distinct channels and assigned an overall confidence score.
How it was made
The dataset is structured into five distinct tables: association links, protein annotations, identifier aliases, amino-acid sequences, and taxonomic metadata. The links table records individual evidence channels alongside an integrated combined score for each protein pair. Sequences are derived from FASTA records, while alias mappings connect STRING identifiers to external biological databases.
Data records are assigned to training, validation, and test subsets using a deterministic CRC32 hash based on entity-specific keys, such as concatenated protein identifier pairs for edges or taxon identifiers for species records. The default partition ratio allocates 98 percent of records to training, 1 percent to validation, and 1 percent to testing.
Key facts
- Version
- 12.0
- Publisher
- LiteFold
- Size category
- 100M to 1B records
- Data modalities
- Tabular, text
- Configurations
- protein_links, protein_info, protein_aliases, protein_sequences, species
- Split ratio
- 98% train, 1% validation, 1% test
- Split method
- Deterministic CRC32 hash on entity keys
What you can use it for
- Interactome network analysis
- Examining whole-organism protein-protein association networks and isolating functional modules using channel-specific or combined confidence scores.
- Graph machine learning
- Training and evaluating graph neural networks for edge prediction or node classification using pre-established, hash-partitioned data splits.
- Sequence-to-interaction modelling
- Connecting primary amino-acid sequences directly with experimental and functional interaction endpoints across diverse species.
- Cross-database identifier mapping
- Translating external database accessions and gene aliases into canonical STRING protein identifiers.
Explore the data
The schema, real rows, full-text search and a SQL filter — read in your browser, nothing downloaded.
Explore
Read the real rows without downloading anything
Reading rows…
LiveRead from LiteFold/STRING at the moment you asked. Nothing is cached or stored — every row above came from that request.
Strengths and limits
Good at
- Integrates multiple distinct evidence channels, spanning direct experiments, computational predictions, curated databases, and text mining.
- Provides deterministic, reproducible train, validation, and test partitions across all constituent tables.
- Pairs extensive network topology directly with corresponding protein amino-acid sequences and taxonomic classifications.
Watch out for
- The volume of association links exceeds one hundred million rows, requiring streaming access or dedicated memory management for large-scale processing.
- Interaction scores combine diverse evidence channels that differ in coverage and reliability across less-studied organisms.
Questions
Can I look inside STRING without downloading it?
Yes. This page reads its schema, real rows and per-column statistics in the browser, with full-text search and a SQL filter. Nothing is downloaded to your machine.
What licence is STRING under?
Its authors declare cc-by-4.0. Read the terms on the original page before reusing the data.
What information is captured in the interaction links table?
The protein links table includes pairwise protein identifiers, individual scores from each underlying evidence channel, and an overall combined confidence score.
How are the training, validation, and test splits constructed?
Splits are determined using a CRC32 hash applied to configuration-specific keys, such as the concatenated pair of protein IDs for links, following a default 98/1/1 percentage distribution.
Can low-confidence protein interactions be filtered out during preparation?
Yes, preparation scripts support setting a minimum threshold for the combined score, allowing users to retain only higher-confidence interactions such as those scoring 700 or above.
Are protein sequences included alongside the interaction network?
Yes, complete amino-acid sequences parsed from source FASTA files are provided in a dedicated configuration alongside protein metadata and sizes.
How to cite it
Szklarczyk, Damian, Rebecca Kirsch, Mikaela Koutrouli, Katerina Nastou, Farrokh Mehryary, Radja Hachilif, Annika L. Gable, et al. "The STRING database in 2023: protein-protein association networks and functional enrichment analyses for any sequenced genome of interest." Nucleic Acids Research 51, no. D1 (2023): D638–D646. https://doi.org/10.1093/nar/gkac1000
Written from STRING's own dataset card. Statements citing a figure that does not appear there are removed before publication. Licence and running details come from our registry.