# UniProt Family & Domains vs. InterPro: Which Database Should You Use UniProt tells you what is already known about a protein, including a curated Family & Domains section on each entry, while InterPro tells you what computational signatures match a sequence by integrating a dozen member databases such as Pfam, PROSITE, and PANTHER. The rule of thumb is simple. If your protein already exists in UniProtKB, start there and read the Family & Domains section, the cross-references, and the evidence tags. If you need to classify a new or unusual sequence, compare signature methods, or annotate thousands of proteins in bulk, go to InterPro and run InterProScan. Most real projects use both, in that order, and knowing which resource answers which question is the difference between a defensible domain annotation and a pile of overlapping signatures you cannot reconcile. ## Quick Answer: Uniprot vs InterPro for Domain Analysis ![Decision path asking whether the protein is already in UniProtKB, whether you need member-database detail, and whether you are working at scale.](/article-images/illustrations/uniprot-vs-interpro-for-domain-analysis-resource-choice.webp) The difference between UniProt and InterPro for domain analysis comes down to their roles in the annotation ecosystem. UniProt is a knowledgebase. Its curators add predicted and experimental functional information to protein sequences stored in the public database, and protein signatures are one of the tools used to classify proteins into families and domains [1]. InterPro is an integrated documentation resource for protein families, domains, and functional sites that was created to merge the major protein signature databases under one roof, with signatures manually integrated into curated InterPro entries that carry abstracts, Gene Ontology mappings, and links to specialized databases [2]. In practice, InterPro provides a classification of UniProtKB sequences [1]. Every InterPro entry links to matched UniProtKB proteins, and InterPro has always provided those matches both on the website and in match XML files on its FTP site [3]. So the two resources are not competitors. They are two views of the same underlying problem, which is turning raw sequences into functional statements. | Criterion | UniProt (Family & Domains) | InterPro | |--|--|--| | Core role | Curated protein knowledgebase with annotated domain features per entry | Integrated classification of families, domains, repeats, and sites from member databases | | Domain data source | Curator-reviewed entries plus automatic annotation, with cross-references to member databases | Direct matches from PROSITE, Pfam, PRINTS, ProDom, SMART, TIGRFAMs, PIRSF, SUPERFAMILY, Gene3D, PANTHER and others [3][2] | | What you see | Domain positions on a specific protein with evidence tags, plus annotated regions, motifs, repeats, PTMs, and variants | Consensus InterPro entries plus the raw member-database matches underneath them | | Coverage philosophy | Deep and reliable for reviewed entries, variable for unreviewed ones | Broad, computational, updated in large releases. Release 13.0 held more than 13,000 entries covering over 78% of UniProtKB [3] | | Best for | Reading what is known about a specific protein, harvesting keywords and function for datasets [4], retrieving clean proteome sets | Running signature scans, classifying novel sequences, building machine-learning features [5], bulk match downloads [3] | | Search modes | Text, sequence similarity, field-constrained queries, REST API | Text and sequence searches on the web server, InterProScan locally or as a web service [3][2] | | Structure of results | One entry per protein, human-readable function, per-feature evidence | One entry per family or domain, many-to-many mapping to proteins [3] | | Weakness | Annotation quality varies between reviewed and unreviewed sections, and propagated annotation can be wrong [6] | Signatures disagree in boundaries and granularity, and a match is a prediction, not a fact | The answer to "uniprot or interpro for domain analysis" is therefore a decision about your starting point. Start at UniProt when the protein is known and you want the distilled, curated view. Start at InterPro when the protein is new, when you need the member-database evidence, or when you are working at scale. ## What Each Resource Actually Is ### UniProt Knowledgebase and the Family & Domains Section The UniProt Knowledgebase, usually written UniProtKB, is where protein sequences live alongside curated functional information. Sequences from genome projects arrive as raw data, and curators add predicted and experimental functional information to them [1]. UniProtKB has two sections. Swiss-Prot holds reviewed entries, checked by human curators, while TrEMBL holds unreviewed, computationally annotated entries. Every protein in TrEMBL eventually has its features computed and propagated, but only reviewed entries carry the full weight of manual curation. The Family & Domains section of a UniProt entry is where domain-level knowledge is summarized for that one protein. You will find named domains with residue ranges, such as "Domain: ETS 279 to 359", alongside regions, repeats, coiled coils, compositionally biased regions, motifs, and sites including active sites and binding sites. Each feature carries an evidence tag. Experimental evidence such as a mutagenesis study or a crystal structure is labeled differently from evidence by similarity or evidence from a computational rule. This distinction matters more than almost anything else in the section, because a domain assignment backed by an experiment and one backed by an automatic annotation pipeline carry very different confidence. UniProt also maintains a controlled vocabulary of keywords, descriptors assigned to protein sequences either through conserved domains or through similarity with already annotated sequences [4]. Those keywords are what make UniProt useful as a filterable, queryable functional database, and they are one of the strongest reasons to retrieve annotations from UniProt rather than reconstructing them yourself. UniProt additionally serves as the [connective tissue](/knowledge/veterinary-medicine/veterinary-anatomy-physiology/connective-tissue-types-loose-dense-and-more) between resources. Its entries carry cross-references to Pfam, SMART, PROSITE, CATH, Gene3D, SUPERFAMILY, and PANTHER, which means the Family & Domains section is effectively a hand-linked index into InterPro's member databases. Many researchers never notice that this cross-reference layer exists, and it is often the fastest way to move from a UniProt entry to the underlying signature evidence. ### InterPro and Its Member Databases InterPro integrates the major protein signature databases. The member set includes PROSITE, PRINTS, ProDom, Pfam, SMART, TIGRFAMs, PIRSF, SUPERFAMILY, Gene3D, and PANTHER, with the last two integrated in later releases [3]. These member databases use different methods. PROSITE builds patterns and profiles. Pfam builds hidden Markov models. PRINTS uses fingerprint groups of motifs. SUPERFAMILY and Gene3D work from structural classifications. PANTHER and TIGRFAMs organize families with functional and phylogenetic intent. A single sequence can match several of these views, and each view has different sensitivity, different granularity, and different biases. InterPro curators manually integrate these signatures into InterPro entries, which provide biological and functional information in the form of an abstract, Gene Ontology mappings, taxonomic range information, protein 3D structure data, and links to specialized databases [2]. When two member databases agree that a region is a catalytic kinase domain, InterPro gives you one merged entry for that domain plus the individual matches beneath it. When they disagree on boundaries, you see that too, which is exactly the information you need when boundary precision matters. The historical release numbers give a sense of scale and of what InterPro classifies beyond plain domains. Release 8.0 contained 11,007 entries, representing 2,573 domains, 8,166 families, 201 repeats, 26 active sites, 21 binding sites, and 20 post-translational modification sites [2]. Release 13.0 contained more than 13,000 entries and covered over 78% of all proteins in UniProtKB [3]. Note that families outnumber domains several-fold. InterPro is not only a domain dictionary. It is a full protein classification system, and treating it purely as a domain lookup is one of the most common ways researchers underuse it. ## How to Use UniProt for Domain Work: A Practical Tutorial ![Five ordered steps from searching an accession to checking variants, PTM and structure sections around the domain.](/article-images/illustrations/uniprot-vs-interpro-for-domain-analysis-uniprot-entry-walk.webp) If you are asking how to use UniProt for domain analysis, the workflow is short and predictable. Suppose you are studying human SPI1, which encodes the PU.1 [transcription factor](/knowledge/molecular-biology/transcription-factor) of the ETS family, a regulator of hematopoietic lineage commitment whose ETS domain integrity affects transcriptional programs [7]. You can find its entry, accession P17947, through the UniProt search at uniprot.org. Step one, search by gene name, protein name, or accession, and filter to the organism you care about. The reviewed Swiss-Prot entry for your organism is usually what you want. Step two, open the entry and scroll to the Family & Domains section. Read the domain ranges, the domain repeat counts, and the annotated regions. For PU.1 you will see the ETS DNA-binding domain annotated with its position, which is precisely the region that mutation studies and docking work then zoom in on [7]. Step three, read the evidence tags on every feature. Experimental evidence beats predicted evidence. Step four, follow the cross-references in the Family & Domains panel to Pfam, PROSITE, and the structure-based databases to see what the signature methods say. Step five, check the Variants, PTM, and Structure sections, because domain context changes how you interpret both. A missense variant sitting inside a conserved DNA-binding domain is a different biological statement than the same substitution in a disordered tail. For programmatic work, use the UniProt REST API. A query like the following pulls the domain features in JSON for SPI1: ``` curl "https://rest.uniprot.org/uniprotkb/P17947.json?fields=ft_domain,ft_region,ft_motif,ft_composition" ``` You can also query in TSV format for spreadsheet-friendly output, add `fields=ft_domain,sequence,protein_name,keyword`, and filter whole proteomes with field-constrained queries. The Proteomes API returns complete, non-redundant protein sets for fully sequenced organisms, which is what you want if you are analyzing an entire proteome rather than one protein [1]. A second UniProt skill worth learning is keyword harvesting. Because keywords are assigned through conserved domains or similarity with annotated sequences [4], you can use controlled keyword sets to define the functional space of a proteome and to build relational summaries of what your gene set does. This is the pattern used in work that projects transcription units and splice variants onto proteome functional space using InterPro and UniProt databases together [4], and it works just as well on a 30-gene list as on a 30,000-gene proteome. ## How to Use InterPro for Domain Work InterPro answers a different question. Where UniProt summarizes what is known, InterPro asks which signatures match, and it answers for any sequence you submit, not only proteins curators have touched. The database supports text and sequence searches through its web server, downloads through anonymous FTP, and InterProScan both locally and as a web service [3][2]. For a single sequence, paste it into the sequence search at the InterPro website. You get back matched InterPro entries, the member-database signatures supporting each match, matched positions, Gene Ontology terms, and representative structures. For a proteome, run InterProScan locally. The standard invocation looks like this: ``` interproscan.sh -i proteins.fasta -f TSV -goterms -pa -cpu 16 ``` The TSV output gives one line per signature match, with the protein, the member database, the signature accession, the matched residue range, and the e-value. The `-goterms` flag adds GO mappings and `-pa` adds pathway annotations, which reflect the curated GO and pathway links attached to InterPro entries [2]. For bulk analysis of already-annotated proteins, skip scanning entirely and download the precomputed match files. InterPro distributes match XML files over FTP covering UniProtKB proteins [3], and those files also include matches to proteins in UniParc, the UniProt archive [3]. If your project needs InterPro classifications for all human proteins, downloading the release matches is dramatically faster and more reproducible than re-scanning, and it pins your analysis to a specific InterPro release version. One InterPro feature that deserves more use than it gets is the Domain Architecture view, introduced among the extended protein match views, which shows the domain composition of protein matches in one graphic [2]. When you are comparing paralogs or homologs across species, architecture at a glance tells you more than a table of coordinates. ## The Criteria That Decide Real Choices ### Curation versus computation UniProt's Family & Domains content on reviewed entries is curated, and curators integrate experimental knowledge with computational predictions. InterPro entries are also curated, but the matches they generate against your sequence are purely computational. This asymmetry has consequences. InterProScan will happily assign a domain to a fragment, a pseudogene product, or a mispredicted gene model, because signatures match whatever sequence you give them. UniProt reviewed entries carry context, such as whether a protein is fragmentary or whether function has been experimentally characterized. When the reliability of the statement matters more than the recall of statements, UniProt wins. When you need to characterize what nobody has characterized, InterPro wins, because it provides classification even for proteins that have never been manually reviewed [1]. ### Granularity and disagreement A UniProt domain feature is one line. An InterPro match can be supported by several member signatures that agree or disagree on boundaries. That disagreement is information. If Pfam and SMART agree on a SH2 domain from residue 112 to 195 and Gene3D places it slightly wider, you know roughly where structural uncertainty begins. If you only read the UniProt summary, you inherit one number with no sense of its stability. Conversely, if you need a single authoritative range to report in a paper, the curated UniProt feature is the cleaner citation. ### Functional annotation machinery InterPro signatures are not only descriptive. They are predictive fuel. EnzML, a multi-label classification method, used InterPro sequence signatures to predict Enzyme Commission classes and achieved more than 98% subset accuracy on a standard set of 300,747 proteins with agreeing Swiss-Prot and KEGG EC annotations, and between 87% and 97% subset accuracy when reannotating eight entire proteomes, with results holding even after redundancy reduction with UniRef100, UniRef90, or UniRef50 clusters [5]. An earlier association-rule approach mined relationships between enzyme classes and significant InterPro entries in Swiss-Prot training sets, applied the rules to TrEMBL test entries, and reached correct enzyme classification rates of about 70% for prokaryote datasets and about 80% for eukaryote datasets [8]. Both studies used UniProt as the ground-truth label source and InterPro as the feature space. That division of labor, UniProt supplies labels and InterPro supplies features, is a template you can copy for your own classifiers. ### Error propagation in comparative genomics Every pipeline that infers domain architectures from genome assemblies, gene predictions, and signature scans accumulates errors, and those errors become systematic bias in evolutionary analyses. A fact-checking approach built on genome-wide ortholog sets and high-quality human and yeast annotations examined domain gain and loss events in the predicted proteomes of nine non-human primates and twenty non-yeast fungi as annotated in UniProt and InterPro, and found that domain losses are overestimated ten-fold in the non-human primate proteins and three-fold in the non-yeast fungal proteins [6]. The lesson for anyone doing uniprot vs interpro comparisons in an evolutionary context is that an apparent domain absence can be an annotation failure rather than a biological event, and that apparent domain losses specifically require independent verification before they enter a phylogenetic argument [6]. ### Scale and reproducibility For single proteins, either website works in minutes. For 50,000 proteins, your choices are precomputed InterPro match downloads [3] or a local InterProScan run. Reproducibility demands you record the InterProScan version, the member-database versions, and the UniProt release date, because all of them move independently between releases. A table of matches generated against one release will not regenerate identically a year later, which is a property to plan for, not discover in review. ## Worked Scenarios ### Scenario 1, one protein, one domain, structural work You are analyzing missense mutations in SPI1 and want to dock a ligand into the PU.1 ETS domain. The published workflow mapped domains using InterPro and UniProt, built wild-type and mutant PU.1 models with SWISS-MODEL, and analyzed the structural alignment and displacement of alpha-carbon positions in UCSF tools [7]. The reason both resources appear in one sentence is that they answer sequential questions. UniProt confirms the reviewed domain range and the functional annotation of PU.1 as an ETS-family transcription factor regulating hematopoietic lineage commitment [7]. InterPro and its member signatures confirm and contextualize the domain assignment, and the cross-referenced structures point you toward a docking-capable PDB entry, 8EQG in this study [7]. If you had started from InterProScan output alone, you would have a residue range but no curated statement about what disrupting that range means for transcriptional programs. ### Scenario 2, annotating a de novo transcriptome You assembled a transcriptome for a non-model organism and have tens of thousands of contigs with no annotation. The published workflow for Quercus pubescens leaves performed deep transcriptomic sequencing, de novo assembly producing 96,006 contigs with a mean length of 618 base pairs, and ran sequence similarity analyses against seven public databases including UniProt, RefSeq, Pfam, InterPro, and KEGG, yielding 83,065 transcripts annotated with gene descriptions, conserved protein domains, or gene ontology terms [9]. The practical pattern is to run similarity search against UniProt for names and function, and InterProScan for domains, then merge. UniProt similarity gives you the closest characterized relative. InterProScan gives you domain architecture even when no close relative exists in the database. Neither alone is sufficient for a defensible annotation pipeline, and reporting both, with database versions, is the standard for transcriptome papers. ### Scenario 3, machine learning on protein function You want to predict function for uncharacterized sequences. Use UniProt reviewed entries as your labeled dataset and InterPro signatures as your features. This is exactly the design of EnzML, which used InterPro signatures as a compact attribute space for multi-label prediction of enzymatic function and hit more than 98% subset accuracy on its evaluation set [5], and of the association-rule work that mined rules linking enzyme classes to InterPro entries from Swiss-Prot training data and applied them to TrEMBL [8]. The approach generalizes beyond enzymes. Any property with curated labels in UniProt, subcellular localization, tissue specificity, disease association, can be turned into a supervised learning problem with InterPro signature matches as input features. ### Scenario 4, domain gains and losses across species You are comparing domain architectures between orthologs to find lineage-specific domain loss. Before you trust any event, remember the measured error rates. In an automated fact-checking analysis of domain gain and loss in predicted proteomes annotated through UniProt and InterPro, domain losses were overestimated ten-fold in non-human primates and three-fold in non-yeast fungi [6]. Build a verification step into your pipeline. Check whether the missing domain region corresponds to a genome assembly gap, a truncated gene prediction, or a fragment flag on the UniProt entry. Treat confirmed domain losses as the subset that survives this check, not the subset your scan output implies. ### Scenario 5, visualizing a protein family You have 300 proteins in a family and want to see who shares which domains, repeats, and disordered regions at a glance. ProFeatMap takes a user-defined list of UniProt accession codes, extracts the most important annotated features available for each protein from well-established databases such as UniProt or InterPro, allocates shapes and colors, and sorts the list based on homologous feature content into a publication-quality two-dimensional map [10]. This is the fastest route from an accession list to a figure a reviewer will actually read, and because it draws from both UniProt and InterPro features, it quietly demonstrates how the two resources complement each other at the visualization layer. ### Scenario 6, comparing candidate genes across species You identified learning and memory candidate genes in one species and want to check that their homologs carry the same functional baggage. A published canine study retrieved the proteins from the UniProt database for dog and human, then compared them in terms of protein families, domains, biological processes, molecular functions, and metabolic pathway locations using InterPro, KEGG, QuickGO, and PSORT databases, finding that some proteins perform the same roles across rat, mouse, dog, and human [11]. The pattern generalizes. UniProt to retrieve and confirm identities, InterPro to compare family and domain composition across orthologs, specialist databases for the dimension you care about next. ## Common Mistakes and Limitations **Reading predicted annotation as experimental fact.** UniProt feature lines carry evidence tags for a reason. An automatically annotated domain on a TrEMBL entry has different standing than the same domain marked with experimental evidence on a reviewed entry. Annotation errors have documented downstream consequences in function prediction, drug discovery, and biotechnology, and they measurably distort evolutionary inference, inflating estimated domain losses ten-fold in one benchmark set of primate proteomes [6]. **Treating "no match" as "no domain."** InterPro covered over 78% of UniProtKB proteins at release 13.0 [3], which means roughly a fifth of proteins had no InterPro match at all. Absence of a signature match is a statement about the signature databases, not about the protein. Novel domains exist precisely in that uncovered fraction, and evolutionary studies that equate unmatched regions with domain absence inflate false loss events [6]. **Confusing families, domains, repeats, and sites.** InterPro's 11,007 entries at release 8.0 broke down into 2,573 domains, 8,166 families, 201 repeats, 26 active sites, 21 binding sites, and 20 post-translational modification sites [2]. A family match tells you what a protein is. A domain match tells you what a segment of it does or resembles. A site match tells you where a few residues matter. Pipelines that filter on "InterPro entry present" without checking entry type mix apples and oranges. **Ignoring release versioning.** InterProScan results change between releases as member databases update their models. UniProt entries change as curation proceeds. Any pipeline that cannot state which InterPro release and which UniProt release produced its numbers is not reproducible, and reviewers increasingly ask. **Downloading when you should scan, or scanning when you should download.** Precomputed InterPro match XML files cover UniProtKB and even UniParc proteins [3]. Scanning your own copy of a UniProt proteome with InterProScan produces different, sometimes broader results, because your sequences may differ from the canonical set and your InterProScan version differs from the release. Pick one approach per project, document it, and do not mix outputs. **Merging match tables without resolving conflicts.** A protein with four member-database hits to the same region has one biological domain, not four. Resolve upward to the InterPro entry, then report the InterPro accession, and keep member-database detail in supplementary files. Conversely, do not collapse distinct InterPro entries that happen to share a GO term. **Assuming keyword and domain sets are interchangeable.** UniProt keywords are a controlled vocabulary assigned through conserved domains or similarity [4]. Domain matches are positional statements about sequence segments. Functional-space analyses that project genes onto keyword-defined functional space [4] and analyses that count domain architectures [6] answer different questions and produce different networks. Choose based on whether your question is about function in the abstract or about sequence segments you can point to. **Forgetting that both databases describe the same sequences.** Every InterPro entry lists matched UniProtKB proteins [3], and every UniProt entry cross-references the signature databases. Analyses that treat UniProt annotations and InterPro matches as independent evidence for the same claim are circular, because UniProt domain annotations are themselves informed by protein signatures [1]. If you need two independent lines of evidence, combine a signature match with experimental data, structural data, or a distinct prediction method, not with the other half of the same annotation pipeline. ## Frequently Asked Questions ### Do I need both UniProt and InterPro for domain analysis? Usually yes. UniProt gives you curated function and context for known proteins, while InterPro gives you signature-based classification for any sequence, including novel ones [1]. A typical pipeline retrieves proteins and their curated annotations from UniProt, then runs InterProScan or downloads InterPro matches for domain-level detail, then merges the two with accessions as keys. ### Which database is better for domain boundaries? UniProt reviewed entries give one curated boundary per domain feature with an evidence tag. InterPro shows the boundaries each member database reports, which may disagree. If you need a single citable range, use the UniProt feature. If you need to judge how stable that range is, read the InterPro member matches underneath it. ### Can I download domain annotations in bulk instead of scanning? Yes. InterPro distributes match XML files by FTP that provide matches to UniProtKB proteins, and additional matches to UniParc proteins, for every release [3]. Downloading precomputed matches is faster and more reproducible than re-scanning a UniProt proteome yourself, provided you record the release version. ### What is the difference between a protein family and a protein domain? A family is a whole-protein classification, meaning the protein belongs to a group with shared ancestry or function. A domain is a segment of a protein with a distinct structure or activity, and proteins often contain several domains. InterPro tracks both explicitly, and its release 8.0 contained 8,166 families against 2,573 domains [2], so the two entry types are distinct categories you should query separately. ### How do I run InterProScan on my own sequences? Install InterProScan locally, then run a command such as `interproscan.sh -i proteins.fasta -f TSV -goterms -pa -cpu 16` to scan a FASTA file and output tabular matches with GO and pathway annotations. If local installation is impractical, the tool is also available through the EBI web service [3], and single sequences can be scanned directly on the InterPro website. ### When should I use UniProt instead of InterPro? Use UniProt when the protein is already in the knowledgebase and you want curated function, experimental evidence, controlled keywords, variants, and cross-references in one entry, or when you need clean proteome sets for whole-[genome analysis](/blog/guides/genome-analysis) [1]. Use InterPro when the sequence is uncharacterized, when you need member-database granularity, when you need bulk match files [3], or when you are building predictive features for machine learning [5]. ## Sources 1. [In silico characterization of proteins: UniProt, InterPro and Integr8.](https://pubmed.ncbi.nlm.nih.gov/18219596/) 2. [InterPro, progress and status in 2005.](https://pubmed.ncbi.nlm.nih.gov/15608177/) 3. [New developments in the InterPro database.](https://pubmed.ncbi.nlm.nih.gov/17202162/) 4. [Projection of gene-protein networks to the functional space of the proteome and its application to analysis of organism complexity.](https://pubmed.ncbi.nlm.nih.gov/20158875/) 5. [EnzML: multi-label prediction of enzyme classes using InterPro signatures.](https://pubmed.ncbi.nlm.nih.gov/22533924/) 6. [Real or fake? Measuring the impact of protein annotation errors on estimates of domain gain and loss events.](https://pubmed.ncbi.nlm.nih.gov/37151482/) 7. [Computational Analysis of SPI1 Missense Mutations and ADMET-Guided Molecular Docking of Cinnamic Acid Targeting the PU.1 ETS Domain: Implications for Hematopoietic Dysregulation and Leukemogenesis.](https://pubmed.ncbi.nlm.nih.gov/42196261/) 8. [Association algorithm to mine the rules that govern enzyme definition and to classify protein sequences.](https://pubmed.ncbi.nlm.nih.gov/16776838/) 9. [RNA-seq analysis of Quercus pubescens Leaves: de novo transcriptome assembly, annotation and functional markers development.](https://pubmed.ncbi.nlm.nih.gov/25393112/) 10. [ProFeatMap: a highly customizable tool for 2D feature representation of protein sets.](https://pubmed.ncbi.nlm.nih.gov/36936371/) 11. [Identification of learning and memory genes in canine; promoter investigation and determining the selective pressure.](https://pubmed.ncbi.nlm.nih.gov/25192892/)