IntAct Reference Ingest Guide¶

Source Information¶

InfoRes ID: infores:intact

Description: IntAct is a curated, open data resource of molecular interactions hosted by EMBL-EBI. It stores experimentally determined molecular interactions, primarily protein–protein interactions, derived from literature curation and direct user submissions. IntAct uses detailed controlled vocabularies (e.g., PSI-MI) to describe experimental details such as interaction type, detection method, and participant roles, enabling consistent representation and downstream reuse of fine-grained interaction evidence.

Citations: - del-Toro N, Duesbury M, Koch M, et al. The IntAct molecular interaction database in 2022. Nucleic Acids Res. 2022;50(D1):D648–D653 - {'Hermjakob H, Montecchi-Palazzi L, Bader G, et al. IntAct': 'an open source molecular interaction database. Nucleic Acids Res. 2004;32(Database issue):D452–D455.'}

Data Access Locations: - {'IntAct main site': 'https://www.ebi.ac.uk/intact/'} - {'IntAct Download / FTP or bulk access': 'https://ftp.ebi.ac.uk/pub/databases/intact/current/psimitab/intact.zip'}

Data Provision Mechanisms: file_download, other, api_endpoint

Data Formats: tsv, xml

Data Versioning and Releases: IntAct is released in regular monthly updates: https://www.ebi.ac.uk/legacy-intact/downloads. Each release is identified by a release number and associated with a release date. The ingest documented here is based on IntAct Release 251 - September 2025; last modified on August 28, 2025 and downloaded on November 16, 2025, using psimitab 2.7 format from: https://ftp.ebi.ac.uk/pub/databases/intact/current. See the IntAct FTP page about the release and contents of the FTP directory: https://www.ebi.ac.uk/intact/download/ftp.

Ingest Information¶

Ingest Categories: primary_knowledge_provider

Utility: IntAct is a key curated source of experimentally supported molecular interactions, particularly protein–protein interactions, which are critical for many NCATS Translator use cases including pathway and network-based reasoning, target discovery, mechanism-of-action hypotheses, and evaluation of biological plausibility for drug–disease or gene–disease connections. It provides detailed interaction-level metadata (e.g., detection method, interaction type, source database, publications) that supports evidence tracking and assessment of interaction quality and context.

Scope: IntAct ingest focuses on experimentally determined molecular interactions where at least one participant can be normalized to a Biolink-compatible protein identifier, with an emphasis on human protein–protein interactions. All interaction records with valid PSI-MI interaction types and detection methods are considered within scope. The ingest is designed to preserve interaction evidence metadata (detection method, interaction type, source database, publications) while providing a Biolink-compliant edge representation.

Relevant Files¶

File Name	Location	Description
intact.zip	https://ftp.ebi.ac.uk/pub/databases/intact/current/psimitab/	MITAB file containing binary molecular interaction records from IntAct for the selected release and organism/filtering configuration used in this ingest. The compressed file has both .txt and negative.txt files; only the positive interactions from the intact.txt file are ingested.

Included Content¶

File Name	Included Records	Fields Used
intact.txt	Human–human protein positive interaction records only. A record is included if both interactors have organism identifiers containing "taxid:9606". These represent experimentally observed molecular interactions curated by IntAct/IMEx, encoded in MITAB 2.7 format	idA, idB (primary identifiers), altIdsA, altIdsB (alternative identifiers), aliasesA, aliasesB (for extracting gene names), taxidA, taxidB (organism filtering), interactionTypes (for predicate mapping), publicationIDs (for evidence), confidenceScores (for edge properties), interactionDetectionMethod (for edge properties)

Future Content Considerations¶

edge_content: Consider expanding beyond human-only interactions to include selected non-human organisms (e.g., key model organisms) where this adds substantial value for mechanistic or translational reasoning, while still maintaining manageable graph size. - Relevant files: intact.txt

edge_content: Consider importing the long tail of records with additional interaction types not taken in first iteration (see excluded_content above). - Relevant files: intact.txt

edge_property_content: Short term, go back and pull edge properties including the Confidence Value(s) (specifically the intact-miscore - we can create a temp intact_confidence_value_miscore in attributes.yaml until we implement standard modeling for this info) and Interaction detection method(s) (use biolink supporting_method_types). Longer term, we can create ExperimentalStudyResult objects to capture metadata for each experiment that supports an edge (e.g. interactor roles, host organism, stoichiometries, features, parameters, experiment-associated publications etc). - Relevant files: intact.txt

edge_content: Revisit retrieval-source modeling of IntAct. This ingest treats IntAct as an interpretive aggregator (primary_knowledge_source) with reported source databases as supporting_data_sources, on the basis that IntAct reviews, vets, and scores imported records. An alternative treats IntAct as a regurgitative aggregator - crediting the reported source database as primary_knowledge_source and IntAct as an aggregator_knowledge_source - plus a new RetrievalSource role (e.g. scorer/evaluator) to record that IntAct assigned the confidence score. Reconsider if same-looking edges carrying IntAct as PKS but different miscores prove confusing to users. - Relevant files: intact.txt

edge_content: Refine predicate/qualifier specificity. Iteration 1 collapses 'direct interaction' (MI:0407) into the generic biolink:physically_interacts_with. A future iteration should map it to biolink:directly_physically_interacts_with, and apply the predicate/qualifier patterns already worked out for the currently-filtered interaction types: phosphorylation/methylation/cleavage/glycosylation reactions -> biolink:affects with a causes_increased_* qualifier; colocalization (MI:0403) and proximity (MI:2364) -> biolink:colocalizes_with (noting MI:0403 is subcellular-fraction co-occurrence, and MI:2364 is a higher-confidence child); association (MI:0914) -> biolink:interacts_with. - Relevant files: intact.txt

Additional Notes: Exploratory data analysis (EDA) of the IntAct MITAB dataset revealed 59 unique PSI-MI interaction types across the full resource, of which 41 appear in the human-only subset for this ingest. A similar 'long tail' distribution was observed for detection methods and source databases, with a small number of highly frequent, widely used terms (e.g., association, physical association, self interaction) and many low-frequency or highly specific terms. Based on these findings, the ingest strategy retains all positive human–human interactions and preserves all observed interaction types, detection methods, and source database annotations (without removing rare terms), while constraining biological scope to human-only records. Negative interactions were excluded for this version of the ingest, as they are not currently used in Translator molecular interaction reasoning or Biolink predicate mapping.

Target Information¶

Edge Types¶

Subject Categories	Predicate	Object Categories	Knowledge Level	Agent Type	UI Explanation
biolink:Protein, biolink:Gene		biolink:Protein, biolink:Gene	knowledge_assertion	manual_agent	Source Intact data aggregates protein-protein interaction statements of different types (e.g. 'physical association', 'direct interaction', 'self interaction') from external manually curated knowledgebases. An automated algorithm assigns a 'Confidence Value' ('intact-miscore') to each unique protein-protein pair across aggregated records. In performing this aggregation and 'meta-assessment', Intact makes its own implicit claims about each interaction, which are captured in Translator edges. The Intact record(s) used to create this particular edge have an 'interaction_type' that asserts or implies a physical interaction between the two proteins ('association', 'physical association', 'putative self interaction', 'self interaction', `direct interaction`). This relationship is represented using the Biolink 'physically interacts with' predicate - which covers direct and indirect physical interactions" Each edge in the UI represents a merged set of IntAct experiment records describing the same molecular interaction. The edge displays a list of supporting publications. All data sources contributing to the merged edge are reported as "supporting data sources" in the biolink:RetrievalSource object attached to the edge.

Node Types¶

Node Category	Source Identifier Types	Additional Notes
biolink:Protein	uniprot	Protein nodes are created when the primary or alternative identifiers include UniProtKB entries. Confirm handling of peptide interactors seen in the data: verify whether they carry UniProtKB identifiers and normalize to biolink:Protein, or whether they represent a coverage gap requiring a separate category/ handling.
biolink:Gene	ensembl, ncbigene, refseq	Gene nodes are created when the primary or alternative identifiers include Ensembl, NCBIGene (Entrez), or RefSeq entries, but no UniProtKB identifier is found.

Provenance Information¶

Contributors: - {'Shilpa Sundar': 'ingest owner, code author'} - {'Matthew Brush, Sierra Moxon': 'data modeling, domain expertise'}

Artifacts: - {'Ingest Survey': 'https://docs.google.com/spreadsheets/d/1sgiHZ_d_o7FWDxhfaP72BVg0FPCNJ7Vx9N6IBuEzHbs/edit?gid=0#gid=0'} - {'Causal Mechanism Mappings': 'https://docs.google.com/spreadsheets/d/1DeAE04O1mz3R9s3dCZpG2hQp9hwif5WjdkUMsBci-u8/edit?gid=1886396936#gid=1886396936'} - {'IntAct EDA Summary': 'https://docs.google.com/spreadsheets/d/1-WF7veY_UXXRogq8f7uRMpgYgiv6lJdx/edit?gid=2068335761#gid=2068335761'} - {'Modeling Notes': 'https://docs.google.com/document/d/1-2sPZEkL6j7JoyX8g2Dk-xSL09qDqaOw1K8tQ_a2oWo/edit?tab=t.0'}