Interaction - Methods summary

The Interaction resource presents protein-protein interaction networks for 16335 genes based on experimental protein-protein interaction data from IntAct, BioGRID, BioPlex and OpenCell, as well as predictions from the deep learning model RosettaFold2-PPI. The networks has been integrated with data related to protein expression, location and classification from the Human Protein Atlas. In addition AlphaFold 3 predicted 3D structures for the interactions present in more than one dataset (consensus) are displayed. The IntAct database relies on experimental molecular interaction data derived from literature curation or direct user submissions, the BioGRID contains curated data derived from publications, BioPlex contributes with interactions based on AP-MS experiments in HEK293T and HCT116 cells and OpenCell provides interaction data based on IP-MS experiments in CRISPR-edited HEK293T cells.

What can you learn from the Interaction Resource?

Learn about:

• the interaction partners of proteins
• the predicted 3D structures of consensus interactions
• the predicted and subcellular location of the proteins in the network
• the expression specificity of the proteins in the network
• the interaction partners expressed in the same single cell type clusters
• known heterodimers from ComplexPortal

Data overview

How has the data been generated?

Experimentally verified interactions included are direct interaction and physical associations with high and medium confidence from IntAct version 2026-01-09, physical multivalidated interactions from BioGRID version 5.0.262, interactions with>75% probability from BioPlex version HEK293T-3.0 and significant physical interactions from OpenCell. The predicted data from the deep-learning model RosettaFold2-PPI is based on the 17,849 interactions with 90% expected precision from the publication Zhang J et al. (2025).

The consensus network plot consists of interaction pairs present in more than one of the datasets. To build the consensus set references for each interaction are compared across the datasets, and an interaction is kept only if the datasets have at least one reference that does not overlap. For BioPlex and OpenCell, we use the database citation as a reference, and if multiple publications are listed an overlap with any of them disqualifies that interaction from being part of the consensus set regarding that database.

The predicted 3D protein structures have been generated in-house based on the AlphaFold 3 source code developed by Deepmind. The structure predictions were run on National Supercomputer Centre (NSC) at Linköping University.

The AI-system Alphafold is a machine learning approach in which the primary amino acid sequence and aligned sequences of homologues together with physical and biological knowledge about related protein structures are incorporated into the design of a deep learning algorithm to directly predict the 3D structure of a protein. Structure predictions have been made for consensus interaction pairs based on protein isoforms preferably mapped to UniProt identifiers, with a length between 6 aa and 4000 aa and containing exclusively standard amino acids.

All structures are displayed using the NGL Viewer.

Figure 1. Genes grouped by the number of first level consensus interaction partners

What is presented in the resource?

This resource presents interaction data for in total 15294 genes based on interactions for 12175 genes from IntAct, 11770 genes from BioGRID, 6374 genes from BioPlex, 4654 genes from OpenCell and 9540 genes from RosettaFold2-PPI. The protein-protein interaction data on the gene pages is displayed as networks with nodes representing interaction partners and edges representing the number of datasets in which the interaction is present. In the RosettaFold2-PPI network and the Aggregated network edges with dashed lines represent the interactions only predicted by RosettaFold2-PPI. The nodes can be coloured according to subcellular location based on data in the Subcellular resource, predicted location based on signal peptide and transmembrane region predictions, single cell type specificity based on RNA expression profiles, single cell cluster or proteinclass, by using the highlight bar in the top of the plot. For genes classified as single cell type enriched or group enriched there is also an option to highlight interactors expressed in the same cell type. There is also an option for custom highlighting of nodes using the top left Filter option in which a query can be built to for example label all nodes that are tissue enriched in both human and mouse brain or belongs to a certain tissue expression cluster.

The predicted 3D protein structures of most consensus interactions can be explored by clicking on the purple dots on the edges in the interaction networks and are displayed together with the corresponding PAE plots and predicted template modelling scores. Known heterodimers sourced from ComplexPortal are highlighted by blue squares on the edges in the interaction networks.

Key publications

Abramson J et al., Accurate structure prediction of biomolecular interactions with AlphaFold 3. Nature. (2024)
PubMed: 38718835 DOI: 10.1038/s41586-024-07487-w

Orchard S et al., The MIntAct project--IntAct as a common curation platform for 11 molecular interaction databases. Nucleic Acids Res. (2014)
PubMed: 24234451 DOI: 10.1093/nar/gkt1115

Oughtred R et al., The BioGRID database: A comprehensive biomedical resource of curated protein, genetic, and chemical interactions. Protein Sci. (2021)
PubMed: 33070389 DOI: 10.1002/pro.3978

Huttlin EL et al., The BioPlex Network: A Systematic Exploration of the Human Interactome. Cell. (2015)
PubMed: 26186194 DOI: 10.1016/j.cell.2015.06.043

Cho NH et al., OpenCell: Endogenous tagging for the cartography of human cellular organization. Science. (2022)
PubMed: 35271311 DOI: 10.1126/science.abi6983

Zhang J et al., Predicting protein-protein interactions in the human proteome. Science. (2025)
PubMed: 40997207 DOI: 10.1126/science.adt1630