# UniProt Interaction Data Not Showing? The most common cause of missing interaction data in UniProt is simple: the protein you are looking at has no curated binary interactions in the underlying molecular interaction databases, so UniProt displays nothing. UniProt does not predict or generate interaction data itself. It cross-references experimentally determined, manually curated interactions from resources such as IntAct and then displays them in the Interaction section of a UniProtKB entry. If the source database has no entries for your protein, or if you are looking at the wrong record type, the section will be empty or absent. The first thing to check is whether you are viewing a reviewed UniProtKB (Swiss-Prot) entry for the correct canonical accession, and whether that entry contains any Interaction content at all in the flat file format. You can do this in one command: ``` curl -s https://rest.uniprot.org/uniprotkb/P04637.txt | grep "Interaction" ``` If that returns nothing for a protein you know has published interactions, the problem is on your side of the pipeline, in the record type, in the query, or in the download format. If it returns nothing for a poorly studied protein, the data may genuinely not exist yet. This article walks through every failure mode in order of how often it occurs, explains the mechanism behind each one, and gives you a repeatable diagnostic sequence plus prevention practices for pipelines that consume UniProt interaction data. ## How Interaction Data Actually Gets Into UniProt Understanding the data flow makes most "missing data" problems obvious. UniProt is not a primary interaction database. Molecular interactions are experimentally determined by laboratories, deposited and curated by members of the IMEx consortium (the International Molecular Exchange consortium, which includes databases such as IntAct, MINT, DIP, and MatrixDB), and UniProt then imports and displays a subset of that curated interaction information. UniProt's curators and automated pipelines propagate binary interactions from these partner resources into the Interaction section of UniProtKB entries, and UniProt also maintains cross-references (DR lines in flat file format) pointing back to the source databases, including IntAct and STRING. This design has three consequences that explain nearly every missing-data complaint: 1. **If the partner database has no curated interaction for the protein, UniProt shows none.** UniProt adds value by unifying and displaying data, not by curating every published pair-bond experiment itself. 2. **The display is a downstream view.** Changes, delays, or filtering rules at the source database or in the propagation pipeline affect what you see on the UniProt website. 3. **Different record types carry different data.** Reviewed Swiss-Prot entries carry the rich annotation, including interactions. Unreviewed TrEMBL entries, UniParc entries (sequence-only archives), and UniRef clusters (sequence clusters) do not display an Interaction section. Because of this, "UniProt interaction data missing" almost always reduces to one of four situations: the data does not exist in the curated sources, you are querying the wrong record, the data exists but your retrieval method dropped it, or the display layer is hiding it. ## Quick Diagnostic Sequence Run through this sequence before anything else. It resolves the majority of cases in under five minutes. 1. **Confirm the accession.** Copy the exact UniProtKB accession (format: six characters, for example P04637) from a reliable source, not a gene name. Gene symbols such as TP53 map to many entries across species and isoforms. 2. **View the flat file, not the web page.** Open `https://rest.uniprot.org/uniprotkb/P04637.txt` in your browser. The text format shows everything in the entry with no collapsing, tabs, or JavaScript. 3. **Check the reviewed status.** In the flat file header, the ID line and the presence of `reviewed:true` in the web URL tell you whether this is Swiss-Prot or TrEMBL. Interaction sections are richest in Swiss-Prot entries. 4. **Search the flat file for Interaction lines and DR lines.** Look for lines beginning with `CC -!- INTERACTION` and cross-reference lines such as `DR IntAct;`. If both are absent, UniProt has no interaction data for this entry. 5. **Check the source database directly.** Search the same accession at IntAct (the EMBL-EBI interaction database that UniProt propagates most binary interactions from). If IntAct has interactions and UniProt does not, note both entry versions and compare again after the next UniProt release, since propagation is not instant. 6. **Rule out record confusion.** Make sure your identifier resolved to UniProtKB and not UniParc, UniRef, or a proteome-wide accession set. Only UniProtKB entries display interaction sections. If all six checks come back negative, the protein most likely has no curated interactions, and the fix is to search interaction databases and prediction resources directly rather than waiting for UniProt to gain the data. ## Troubleshooting Table Use this table as the core diagnostic reference. Symptoms are ordered from most common to rarest. | Symptom | Likely cause | How to confirm | Fix | |--|--|--|--| | Interaction section completely absent from the entry page | No curated binary interactions exist for this protein in the source databases | Check IntAct and BioGRID directly with the same accession; if they are empty too, the data does not exist | Query the primary interaction databases directly, use prediction resources like STRING for hypotheses, or wait for curation of new literature | | Entry has an IntAct cross-reference (DR line) but no Interaction section | Interaction data is held at the partner level and only cross-referenced, or the propagation has not completed | Open the DR IntAct link from the flat file and inspect the partner database record | Use the cross-reference to pull interactions from the partner database itself | | Interaction section visible on one accession of a gene but not another | Isoform-level annotation, or you are on a TrEMBL entry representing another isoform | Compare the `DR` and `CC` interaction lines across all mapped accessions for the gene | Anchor your analysis to the canonical Swiss-Prot accession and track isoform-specific annotations explicitly | | Interaction column missing from a downloaded TSV or spreadsheet | Default download fields do not include interaction or cross-reference columns | Reopen the download dialog and inspect the selected columns | Add the Interaction and cross-reference columns explicitly, or request the needed fields in an API query | | API response contains no interaction fields | The query did not request interaction-related fields, or the endpoint used returns a minimal record | Compare the API response against the flat file for the same accession | Add the relevant cross-reference and comment fields to the `fields` parameter of the request | | Interaction viewer or section fails to load on the website | JavaScript blocked, extension interference, or a transient cache or CDN problem | Load the flat file version, which bypasses all front-end rendering | Disable blockers for the domain, clear cache, or use the text and XML formats for programmatic work | | Interactions shown previously have disappeared | Accession merged or demerged, entry version updated, or an interaction was retracted or re-annotated upstream | Check the entry version history (UniSave) and look for a secondary accession mapping | Follow the mapping to the current accession, and pin accessions and versions in your analysis | | Protein found by name search shows no data, but the accession does | Name search resolved to the wrong entry, wrong species, or a TrEMBL homolog | Note the accession returned by the search and compare it to your target | Search by accession, or restrict search results to the target organism and reviewed entries | | Interactions exist in literature but nowhere in UniProt | Published interactions have not been curated into the partner databases yet | Search PubMed for the interaction and check curation status at the partner databases | Interactions cannot be forced into UniProt, use the primary literature or submit to curation | | Interaction data expected from a prediction tool is absent | Prediction resources such as STRING supply scores, not curated binary interactions, and are treated differently | Distinguish curated evidence (IMEx databases) from predicted evidence (STRING-style co-expression or genomic proximity scores) | Use STRING directly for predictions and interpret them as hypotheses, not curated facts | ## Causes, From Most Common to Rarest ### 1. The protein genuinely has no curated interactions This is the single most frequent reason. UniProtKB contains hundreds of thousands of reviewed entries, but curation effort concentrates on well-studied model organisms and medically important proteins. A protein from a less-studied organism, an uncharacterized open reading frame, or a newly described gene may have zero experimentally determined binary interactions in any public database. In that case the Interaction section on the UniProt entry is absent, and nothing is broken. The mechanism is straightforward. Interaction databases like IntAct curate interactions from peer-reviewed publications, encoding each binary pair with evidence, experimental method (often using PSI-MI vocabulary, the Proteomics Standards Initiative molecular interaction terminology), and participant details. UniProt then imports a curated view of those binary interactions into its entries. If no publication describing a physical interaction with your protein has been curated, there is nothing to import. How to confirm and what to do: - Search the accession at IntAct and BioGRID. If both return no interactions, the curated record is genuinely empty. - Search PubMed for the protein name plus terms such as "interaction," "co-immunoprecipitation," "yeast two-hybrid," "pull-down," or "AP-MS" (affinity purification followed by mass spectrometry). If papers report interactions, the gap is in curation, not in biology. - For hypothesized interactions, use STRING, which aggregates predicted and evidence-weighted associations including genomic context, co-expression, and text mining. Treat STRING results as computational hypotheses, distinct from the experimentally curated data UniProt displays. - Consider submitting well-documented interactions to an IMEx database for curation if your own experiments produced them. This is how the public interaction corpus grows. A practical habit: always test your retrieval pipeline on a protein with dense annotation, such as human TP53 (accession P04637), which carries a large number of curated interactions and cross-references. If your pipeline returns interaction data for P04637 and nothing for your protein of interest, your pipeline works and the target protein is the issue. ### 2. You are looking at an unreviewed TrEMBL entry UniProtKB consists of two sections. Swiss-Prot entries are manually annotated and reviewed by curators. TrEMBL entries are computationally annotated and unreviewed, and they carry far less manual content. Interaction sections with rich binary interaction descriptions appear primarily on reviewed entries. If a gene symbol search or an automated identifier mapping lands you on a TrEMBL entry for a paralog, an isoform, or a fragment, the Interaction section will be missing or sparse even though a well-annotated reviewed entry for the same gene exists elsewhere. Confirm this by checking the entry header. Reviewed entries are searchable and filterable, and in the web interface the reviewed status is displayed on the entry and in search results. In the flat file, the entry name line and accession pattern give clues, and the search filters make the distinction explicit. Fix: search by accession rather than name whenever possible, or restrict search results to reviewed entries before picking a record. When a gene symbol resolves ambiguously, verify organism, gene, and protein name before treating the entry as authoritative. ### 3. You are on the wrong record type: UniParc, UniRef, or a proteome page UniProt serves several databases beyond UniProtKB. UniParc stores raw sequences without biological annotation. UniRef clusters sequences at 100%, 90%, and 50% identity thresholds to speed up searches and reduce redundancy. Proteome pages present the full set of proteins for an organism. None of these record types display an Interaction section, because interactions are properties of annotated UniProtKB entries, not of bare sequences or clusters. This failure mode is common in scripts. A pipeline that resolves an identifier through a cross-database service or a cluster ID can silently end up with a UniRef representative ID and then report "no interaction data found" for an entry that actually has plenty. Confirm by checking that your identifier matches the UniProtKB accession pattern and that the URL you fetch contains `uniprotkb` rather than `uniparc`, `uniref`, or `proteomes`. Fix: normalize all identifiers to canonical UniProtKB accessions at the start of the pipeline, and log any identifier that fails to resolve. ### 4. The web page hides or fails to render the section UniProt entry pages organize content into sections, and some content is collapsed by default or loaded dynamically. The Interaction display, including any graphical interaction viewer, depends on JavaScript and on the browser loading resources from the site and from partner services. Browser extensions that block scripts or third-party requests, aggressive privacy settings, or a stale cached page can make a populated Interaction section appear empty. Confirm by loading the flat file version of the same entry, which contains the same data as plain text with no rendering layer. If the flat file shows `CC -!- INTERACTION` blocks but the web page shows nothing, the problem is browser-side. Fix: allow scripts for the domain, clear the cache, try a different browser, or bypass the front end entirely for programmatic work by using the text, XML, or JSON formats. ### 5. Your download or API query dropped the interaction columns This is the most frequent cause for people who already know the data exists. UniProt lets you choose which columns to download, and the defaults are minimal. A TSV exported from the website, or an API query that returns the default field set, will not include interaction content unless you request it. The result looks exactly like missing data. Fixes: - On the website, open the Columns selector in the results view and add the Interaction section and the cross-reference columns (including IntAct) before downloading. - In the REST API, pass an explicit `fields` parameter listing the fields you need. UniProt documents the available query and result fields, including cross-reference fields for interaction databases, in its API query documentation. Consult that field list rather than guessing field names, because names change between releases and differ between query filters and result columns. - For complete records, fetch the full flat file (.txt), XML, or JSON format instead of a column-limited TSV, then parse the interaction content locally. A worked example that always works for a single accession: ``` curl -s https://rest.uniprot.org/uniprotkb/P04637.txt > p53.txt grep -c "DR IntAct" p53.txt grep "INTERACTION" p53.txt | head ``` For bulk retrieval, search the REST API with your accession list, request the interaction-related fields explicitly, and download in TSV or JSON. Verify the row count against your input list so silent identifier drops do not masquerade as missing interactions. ### 6. Isoform confusion Many human proteins have multiple isoforms, and interaction evidence is frequently isoform-specific. A protein-protein interaction may involve isoform 2 and not the canonical isoform 1, or vice versa. UniProt annotates isoform-level interactions where the evidence supports it, and the partner databases store interaction participants at the isoform level. If you expect an interaction on the canonical sequence and it is annotated on another isoform, or if your identifier mapping resolved to an isoform-specific accession, the data appears "missing" when you look at the wrong sequence. Confirm by reading the Interaction section carefully in the flat file. Isoform-specific interactions are annotated with isoform identifiers, and UniProt's isoform annotation in the entry tells you which sequence each isoform corresponds to. Fix: decide up front whether your analysis is at the gene level, the canonical protein level, or the isoform level, and keep that decision consistent across UniProt, the interaction databases, and your own [downstream processing](/knowledge/molecular-biology/downstream-processing). ### 7. Curation lag and propagation delay Interaction curation is manual and slow. A landmark paper can report a new interaction months or years before a curator at an IMEx database encodes it, and the transfer of that curation into UniProt entries adds a further interval. UniProt releases updated entries on a regular schedule, and data propagated from partner databases moves on those release cycles. A protein that "should" have interactions based on recent literature may correctly show none. Confirm by comparing the current date against the entry's most recent version date (shown on the entry and available through UniSave, UniProt's entry version archive) and against the literature. If the publication predates the last release by a comfortable margin and the interaction is still absent, check the partner database directly, because the gap is usually there rather than in UniProt. Fix: build your workflow to tolerate this lag. Cross-check primary interaction databases, and re-run queries after major UniProt releases if currency matters to your analysis. ### 8. Merged, deleted, or demerged accessions Accessions in UniProt are stable, but entries are not frozen. Two entries can merge into one, a merged entry can demerge into several, and an old accession can become a secondary accession that redirects to a new one. If your dataset contains an accession that was merged since you downloaded it, a naive query can return an error, an empty result, or a redirect you did not anticipate, and the interaction data you remember now lives under the new primary accession. Confirm by checking the entry's "Secondary accession" lines in the flat file, which list prior accessions, and by using UniSave to see version history. Fix: when an accession returns nothing, query the UniProt search interface with the accession string regardless of format, because the search resolves secondary accessions to current entries. In pipelines, log redirects and merges instead of treating them as failures. ### 9. Confusing curated interactions with predicted associations A recurring expectation mismatch: people look for an interaction in UniProt, find nothing, and conclude the data is missing, when in fact the only available evidence for that interaction is computational prediction. UniProt's Interaction section reports curated binary interactions with experimental evidence. Predicted functional associations from STRING, co-expression correlations, or pathway co-membership are different data types that UniProt does not present as curated interaction annotations. Confirm by asking what kind of evidence you expect. If the evidence is a high-confidence STRING score rather than a published binary interaction, UniProt will not show it as an interaction. Fix: use STRING for predicted associations, IntAct and other IMEx databases for curated physical interactions, and UniProt as the unified annotation view over the curated layer. ### 10. Silent identifier problems in bulk workflows The rarest but most damaging variant: a batch mapping or download runs without errors, and the resulting table simply has blank interaction cells for many or all proteins. Causes include identifiers that were never UniProt accessions to begin with (gene names, RefSeq or Ensembl IDs passed without mapping), identifiers from a different species than assumed, obsolete accessions, and truncated identifiers from spreadsheet software that converted accessions to numbers or dates. Fixes that prevent this class of failure: - Use UniProt's ID mapping service to convert any identifier set to UniProtKB accessions before querying, and keep the mapping report so you know which inputs failed. - Pin the organism in every query so a human gene name does not resolve to a mouse TrEMBL entry. - Store accessions as text in spreadsheets to prevent conversion corruption. - After any bulk download, assert that the number of retrieved entries matches the number of mapped accessions and spot-check one entry with dense annotations against the website. ## Worked Example: Verifying Interaction Data End to End ![Calculation showing how to fetch the P04637 flat file, count DR IntAct lines with grep -c, and compare the count against your protein of interest.](/article-images/illustrations/uniprot-interaction-data-missing-positive-control-count.webp) Here is a complete verification for a protein with known rich interaction data, using only UniProt itself. 1. Fetch the flat file: `curl -s https://rest.uniprot.org/uniprotkb/P04637.txt -o P04637.txt` 2. Confirm it is the right record: check the first lines for the entry name and organism (human). 3. Count interaction cross-references: `grep -c "DR IntAct" P04637.txt`. A healthy count confirms the pipeline and the endpoint. 4. List the curated interaction comment blocks: `grep "INTERACTION" P04637.txt`. Each block names a interactor, its accession, and the number of experiments. 5. Pull the interaction partner accessions: `grep -A1 "INTERACTION" P04637.txt | grep -oE "[OPQ][0-9][A-Z0-9]{3}"` to extract UniProtKB-style accessions of partners for downstream queries. 6. Repeat the same fetch with `.json` and confirm the cross-reference data appears there too, so your parser handles the format your pipeline actually consumes. Once this positive control passes, apply the identical commands to your protein of interest. The difference between the two outputs tells you immediately whether the problem is your method or the data. ## A Short UniProt Tutorial: Finding What You Need Because many missing-data complaints trace back to unfamiliarity with the resource, here is a compact orientation for using UniProt effectively. **Searching.** The UniProt homepage search box accepts accessions, entry names, gene names, protein names, sequences, and structured query syntax. For a known protein, search the accession directly. For exploratory work, use the query builder: restrict results to your organism, to reviewed entries, and to the cross-reference or annotation categories you care about. Search results can be filtered interactively, and the filter state is reflected in the URL, which makes searches reproducible and shareable. **Reading an entry.** A UniProtKB entry page is organized into functional sections: function, subcellular location, domain, sequence, interaction, expression, and so on, each with a small icon and header. Within a section, annotation lines carry evidence tags that link to the evidence source. The Interaction section lists binary interactions, usually with the partner's name, its accession, and the number of publications supporting the interaction. Cross-references to IntAct, STRING, and other resources appear in the cross-references area and link out to the source records. **Choosing a format.** The Download or Export controls on the entry and results pages offer flat file text, FASTA, XML, JSON, spreadsheet, and other formats. Use FASTA for sequences only, text for complete human-readable records including interaction comment blocks, XML or JSON for structured parsing, and TSV for tabular bulk work with explicitly chosen columns. **Programmatic access.** UniProt provides a REST API with documented endpoints for searching and retrieving entries, documented query fields, and an ID mapping service. The pattern is stable: build a query, request the fields you need, page through results for large sets, and download in a machine-readable format. For interaction work specifically, the two things to request explicitly are the interaction annotation content and the cross-references to interaction databases. **Versioning.** Every entry has a version number and a modification date, and UniProt archives previous versions through UniSave. Citing or pinning an entry version makes analyses reproducible across release cycles. ## How to Get Interaction Data When UniProt Has None ![Decision path separating curated binary interactions in IMEx databases from predicted associations in STRING and the unified view in UniProt.](/article-images/illustrations/uniprot-interaction-data-missing-which-source-to-use.webp) When the diagnostic sequence confirms that UniProt holds nothing for your protein, escalate to the sources and tools that the field maintains for exactly this purpose. - **IntAct**, maintained at EMBL-EBI, is the primary curated binary interaction database and the main source UniProt propagates interaction data from. Search it with the same accession. Its data model captures experimental methods, evidence, and participant details at a finer grain than UniProt displays. - **BioGRID** curates physical and genetic interactions from the literature with a strong focus on model organisms. It frequently holds interactions for proteins that other databases have not yet covered. - **DIP, MINT, and MatrixDB** are additional IMEx-consortium and related curated databases, each with historical strengths in particular organisms or interaction classes such as extracellular matrix interactions. - **STRING** aggregates predicted and experimental associations into scored networks. Use it when curated data is absent and you need hypotheses, and interpret its scores as confidence in an association rather than proof of a direct physical interaction. - **Primary literature.** For any interaction that matters to your conclusions, go to the original publication. Databases encode a summarized view of experiments, and reading the methods section of the source paper tells you the experimental technique, which is essential for judging evidence quality. A sound workflow retrieves curated interactions from the IMEx databases through their own APIs or download files, uses UniProt for unified protein annotation and cross-references, and reserves prediction tools for generating, not confirming, hypotheses. ## Prevention Practices for Pipelines That Consume UniProt Data Missing-data bugs usually enter pipelines through assumptions. These practices keep them out. - **Pin accessions and versions.** Store canonical UniProtKB accessions and, where reproducibility matters, the entry version and release date alongside your results. - **Always run a positive control.** Include one densely annotated accession in every batch retrieval and assert that interaction content comes back for it. This single check catches endpoint changes, format changes, and field-name changes instantly. - **Request fields explicitly.** Never rely on default column sets in downloads or API responses. List the fields your analysis needs. - **Log redirects and empty results separately.** A merged accession, a TrEMBL entry, and a genuinely interaction-free protein are three different situations that all look like "no data" if you do not log them distinctly. - **Normalize identifiers once, upstream.** Map all input identifiers to UniProtKB accessions through the ID mapping service at the start of the workflow, and carry the mapping report forward. - **Date-stamp your data.** Interaction curation grows continuously. Record the download date so a blank result six months from now can be distinguished from a blank result today. - **Cross-check one source of truth per data type.** Curated binary interactions come from IMEx databases, predictions from STRING, and unified annotation from UniProt. Keeping these separate prevents both false absence and false presence. ## Common Mistakes and Limitations Several mistakes account for most of the frustration in this area. - **Treating UniProt as the primary source of interactions.** It is a unified annotation view over curated data from other databases. When it shows nothing, the first escalation is to the partner databases, not to bug reports. - **Searching by gene name and trusting the first hit.** Gene symbols are ambiguous across species, paralogs, and isoforms. Search by accession, or verify the organism and reviewed status of whatever the name search returns. - **Reading blank TSV cells as evidence of absence.** A missing column is not a missing interaction. Check the flat file for the same accession before concluding anything. - **Mixing interaction types.** Curated physical binary interactions, genetic interactions, and predicted functional associations answer different questions. A blank curated record does not contradict a high predicted score, and neither validates the other. - **Assuming completeness or currency.** Curation lags literature by design. For a specific interaction that matters, the publication itself is the ground truth, and database views converge on it at their own pace. - **Ignoring isoforms.** Interaction evidence attaches to specific sequences. Analyses that collapse everything to a gene symbol will misplace isoform-specific interactions. One limitation applies to this entire domain: interaction databases reflect what has been published and curated, not what is true in every cell type and condition. Absence of a curated interaction is not evidence that two proteins never interact, and presence of one curated interaction does not guarantee it occurs in your experimental context. Individual research questions need evaluation against current literature and experimental validation in the relevant system. ## Frequently Asked Questions ### Why does a UniProt entry show no Interaction section at all? Most often because no curated binary interactions exist for that protein in the source databases UniProt imports from. Check the flat file format for the entry, then search IntAct and BioGRID with the same accession. If those are also empty, the data has not been curated yet and cannot be made to appear in UniProt. ### Does UniProt predict protein interactions? No. UniProt displays experimentally determined, manually curated binary interactions propagated from partner databases in the IMEx consortium. Predicted associations are the domain of tools such as STRING, which you should use directly for prediction-based hypotheses. ### What is the difference between UniProt interaction data and STRING? UniProt's Interaction section reports curated interactions backed by experimental evidence from literature curation. STRING produces scored associations from multiple evidence types including prediction methods. The two answer different questions and should not be treated as interchangeable. ### How do I download all interactions for a list of proteins? Map your identifiers to UniProtKB accessions with the ID mapping service, query the REST API with those accessions, explicitly request the interaction and cross-reference fields, and download in TSV or JSON. Verify the output row count against your mapped input list before analysis. ### Why do interactions appear in IntAct but not on UniProt? Interaction curation at partner databases and propagation into UniProt entries run on different schedules, so a recent curation may not have reached UniProt yet, or the interaction may be represented only as a cross-reference. Use the IntAct cross-reference on the UniProt entry to reach the fuller partner record. ### Why are interactions listed for one isoform but not the canonical sequence? Interaction evidence is often isoform-specific because the experimental construct used a particular isoform. UniProt annotates these at the isoform level where evidence supports it. Read the Interaction section in the flat file carefully and note which isoform identifier each interaction cites. ### How current is UniProt interaction data? Interaction content updates on UniProt release cycles, with curation at partner databases occurring continuously in between. The modification date on each entry shows when it last changed, and UniSave archives previous versions. For the most current view of a specific interaction, check the partner database and the primary literature. ### Why did a protein that used to show interactions stop showing them? The most likely explanations are an accession merge or demerge, an entry version change, or an upstream re-annotation or retraction at the source database. Check the entry version history and secondary accession lines, follow any mapping to the current primary accession, and compare your old record against the current flat file.