How many protein interactions are in the human interactome?
Matic BrozComputational chemist
The human interactome probably contains hundreds of thousands to a few million direct protein-protein interactions. No experiment or database has counted them all.
The clearest measured reference is HuRI, with 52,569 verified binary interactions among 8,275 proteins. Larger database totals answer different questions because they may include repeated evidence, proteins found in the same complex, or predicted functional links.
How many protein-protein interactions are there in humans?
The human binary interactome probably contains roughly 480,000 to 2.6 million protein-protein interactions, but there is no complete census.
HuRI's authors estimate that their 52,569 verified interactions represent 2% to 11% of the full binary interactome.[1] Dividing the measured count by those coverage limits gives about 477,900 to 2,628,450 interactions. The range is calculated rather than directly observed.
An earlier statistical analysis estimated about 650,000 human protein interactions from incomplete network samples.[2] That figure sits inside the HuRI-derived range, but both results depend on assumptions about network sampling and assay-detectable binary interactions.
The estimate does not include every context-specific association among isoforms and modified forms. The human proteome contains far more molecular forms than reference protein types, and an interaction can appear in one cell type or condition but not another.
How much of the human interactome has been mapped?
HuRI maps 52,569 verified binary interactions among 8,275 proteins, and its authors estimate that this covers 2% to 11% of the binary human interactome.[1]
HuRI tested more than 150 million protein pairs drawn from 17,408 protein-coding genes using three versions of yeast two-hybrid screening. Repeated screening, pairwise retesting, sequence confirmation, and checks with orthogonal assays produced the published HI-III-20 map.[1] Its 8,275 connected proteins cover only part of the reference proteome discussed in our guide to protein counts.
BioPlex measures a different slice of the interactome. Affinity-purification mass spectrometry in HEK293T cells produced 118,162 associations among 14,586 proteins. A second map from HCT116 cells contained 70,966 associations among 10,531 proteins, and comparison of the two showed extensive cell-specific differences.[3] These are co-complex or co-purifying associations, so they do not establish that every pair touches directly.
The chart shows protein coverage in systematic experimental maps. Interaction totals are omitted because HuRI screens for binary pairs, while BioPlex maps proteins recovered together in particular cell lines.[1][3]
Why do protein interaction databases report different totals?
Protein interaction databases report different totals because they count different evidence, entities, and duplicates. A database row is not necessarily one unique biological interaction.
| Database | Date or release | Species scope | Evidence type | Proteins or nodes | Interactions or edges | Counting rule |
|---|---|---|---|---|---|---|
| HuRI | HI-III-20 (2020) | Human ORFeome screen | Verified yeast two-hybrid binary interactions | 8,275 | 52,569 | Published verified protein-pair count |
| BioPlex 3.0 | 2021 HEK293T network | Homo sapiens HEK293T cells | Affinity-purification mass-spectrometry associations | 14,586 | 118,162 | Published network pair count from 10,128 affinity purifications |
| STRING | v12.0 (accessed August 9 2026) | Homo sapiens taxon 9606 | Known and predicted functional associations | 19,622 | 6,857,702 | One undirected pair with combined score 150 or higher; duplicate orientation removed |
| BioGRID | 5.0.260 (August 2026) | Homo sapiens organism row, including cross-species records | Curated physical interactions | 29,164 | 1,120,624 | Official non-redundant count: each unordered A-B pair once regardless of method or publication |
| IntAct | Release 252; query accessed August 9 2026 | Human-human protein pairs | Curated or submitted positive interaction evidence | 30,804 | 1,058,685 | One positive binary evidence entry per result after expansion; repeated pair-method-publication evidence retained |
HuRI and BioPlex figures come from their primary publications.[1][3] STRING figures come from the official v12.0 human download, BioGRID figures from its August 2026 organism statistics, and IntAct figures from the stated MIQL query under release 252.[4][6][7][8][9] The same rows and counting rules are available in the downloadable CSV.
STRING is the largest row here because it is a functional association network. It combines direct and indirect links from experiments, curated databases, co-expression, text mining, genomic context, computational prediction, and evidence transferred between organisms.[5] The 6.86 million edges therefore include much more than experimentally confirmed human protein contacts.
BioGRID collapses its curated physical evidence to non-redundant A-B pairs for the table. IntAct's result count keeps evidence entries generated by different methods, publications, and complex expansions. BioGRID's Homo sapiens row can also include cross-species interactions, while the IntAct query requires both interactors to be human proteins.[6][8][9]
What counts as a protein interaction?
A direct binary contact, a shared protein complex, a curated experimental observation, and a predicted functional association are different units and should be reported separately.
HuRI's yeast two-hybrid screens seek pairs that can interact directly in a binary assay. BioPlex captures proteins that co-purify with a tagged bait, which can include direct binding partners and indirect members of the same complex. BioGRID and IntAct preserve evidence from many experimental systems, so the same pair can appear in several publications or methods. STRING also adds indirect and predicted associations.[1][3][5][6][9]
This is why the five database counts cannot be added. Their records overlap, their identifiers differ, and their edges describe different biological claims.
Why is the human interactome still incomplete?
The human interactome remains incomplete because protein interactions depend on cell type, time, location, molecular form, and experimental sensitivity.
HuRI found most of its interactions in only one of nine screens, even though repeat detections were not more precise than single-screen detections. The authors concluded that many interactions may be weak or transient and therefore difficult to detect.[1] Binary assays can also miss interactions that need a particular modification, membrane environment, ligand, or multi-protein assembly.
BioPlex showed that two human cell lines share many core complexes but also contain substantial cell-specific network wiring.[3] A static human interaction network is a growing evidence map assembled across different biological contexts.
How were the database counts calculated?
Each database figure uses a source-specific rule, and none was added to another database total.
HuRI and BioPlex counts are reported directly by their primary papers. For STRING v12.0, the human protein.links file contains 13,715,404 oriented rows with combined scores from 150 to 999. Counting only rows where the first protein identifier sorts before the second removes the duplicate orientation and gives 6,857,702 undirected associations. The resulting network contains 19,622 protein identifiers with at least one edge.[4]
BioGRID build 5.0.260 reports 1,120,624 non-redundant physical interactions and 29,164 unique genes for its Homo sapiens row. BioGRID defines a non-redundant interaction as one unordered A-B pair regardless of publication, experimental system, or direction.[6]
The IntAct total uses the query (taxidA:9606 AND taxidB:9606) AND ptypeA:protein AND ptypeB:protein AND negative:false. On August 9, 2026, the portal returned 1,058,685 interaction table entries and 30,804 interactor accessions.[8][9] These are query-result records, not a deduplicated count of unique human gene pairs.
Sources▼
- A reference map of the human binary protein interactome Nature · 2020. https://pmc.ncbi.nlm.nih.gov/articles/PMC7169983/
- Estimating the size of the human interactome Proceedings of the National Academy of Sciences · 2008. https://pmc.ncbi.nlm.nih.gov/articles/PMC2383957/
- Dual proteome-scale networks reveal cell-specific remodeling of the human interactome Cell · 2021. https://pubmed.ncbi.nlm.nih.gov/33961781/
- STRING v12.0 Homo sapiens downloads STRING Consortium · August 9, 2026. https://string-db.org/cgi/download?species_text=Homo+sapiens
- STRING database content and evidence sources STRING Consortium · August 9, 2026. https://string-db.org/cgi/about?footer_active_subpage=content
- BioGRID database statistics, build 5.0.260 BioGRID · August 9, 2026. https://wiki.thebiogrid.org/doku.php/statistics
- IntAct Molecular Interaction Database, release 252 EMBL-EBI · August 9, 2026. https://www.ebi.ac.uk/intact/
- IntAct human-human positive protein interaction query EMBL-EBI · August 9, 2026. https://www.ebi.ac.uk/intact/search?query=%28taxidA%3A9606%20AND%20taxidB%3A9606%29%20AND%20ptypeA%3Aprotein%20AND%20ptypeB%3Aprotein%20AND%20negative%3Afalse
- IntAct user guide: Advanced MIQL search EMBL-EBI · August 9, 2026. https://www.ebi.ac.uk/intact/documentation/user-guide#advanced_search
Founder and computational chemist, ProteinIQ
Dr. Matic Broz is the founder of ProteinIQ and a computational chemist. He completed a PhD focused on protein structure, molecular dynamics, and neural networks, and writes about structural biology and scientific software.