# How to Read the Function Section of a UniProt Entry The Function section of a UniProt entry is the organized summary of what a protein does, combining a plain-language FUNCTION statement with structured annotations such as catalytic activity, cofactors, subunit composition, pathways, disease associations, and positional features like active sites and binding sites. Every statement and feature carries an evidence tag that tells you whether the claim comes from direct experiment, inference from a characterized homolog, or an automatic prediction. This section is the part of a UniProt entry that most downstream work actually depends on. Variant interpretation, proteomics annotation, pathway construction, and machine-learning function prediction all consume it, and misreading one evidence tag can propagate a wrong assumption through an entire analysis. UniProt now contains over 203 million protein entries, of which only about 2 percent have been experimentally validated, so the gap between what is sequenced and what is known functionally is enormous [1]. Reading the Function section correctly is therefore the skill that separates a careful annotation from a copy-paste citation. ## What Is the Function Section of a UniProt Entry ### Where the Function Section Sits in an Entry When you open an entry at uniprot.org, for example the human tumor suppressor p53 (accession P04637), the page is organized into a left-hand navigation pane with top-level sections: Function, Structure, Interactions, Expression, Alphabetical Index, and Sequence. Clicking Function reveals the subsections in a fixed order that mirrors the flat-file format: Function, catalytic activity, cofactors, activity regulation, biophysicochemical properties, subunit, interaction, cellular location, pathway, post-translational modification, disease, and miscellaneous comments. Each subsection is rendered as a list of statements. Each statement ends with one or more small colored evidence tags. Hovering over a tag reveals the exact source, such as a PubMed identifier for an experimental result, a rule name for a prediction, or a related entry from which the claim was propagated. This tagging system is the single most valuable feature of the section, and it is the thing new users most often overlook. In the underlying flat file, which you can retrieve as plain text, the Function section corresponds to a block of CC (comment) lines preceded by a block of FT (feature table) lines and followed by DR (database cross-reference) lines. The web interface simply visualizes these lines. The ProtVista component, an open-source JavaScript viewer embedded in the UniProt web page, draws the positional features along the sequence so you can see at a glance whether an active site, a domain, and a disease variant overlap [2]. ### Swiss-Prot Versus TrEMBL: Why Curation Status Changes Everything UniProtKB (the UniProt Knowledgebase) contains two distinct parts. Swiss-Prot entries are reviewed by expert curators who read the literature and write the Function section by hand. TrEMBL entries are unreviewed and generated computationally, with most of their content coming from rule-based annotation and similarity propagation rather than human reading. This distinction changes how much weight each Function statement deserves. UniProt has been undergoing a major reorganization of its protein space across releases from 2025_04 through 2026_02, and one explicit goal of that reorganization is to keep all reviewed Swiss-Prot entries along with proteins from selected reference proteomes and unreviewed entries carrying biologically important data, while pruning redundant unreviewed records. Analysis of UniProtKB and UniParc records accessed in January 2026 showed roughly 200 million entries remained in UniProtKB, with about 58.9 million designated for removal in release 2026_02 [3]. For a reader, the practical lesson is stable: a reviewed Swiss-Prot Function section is the closest thing UniProt offers to a gold standard, and an unreviewed TrEMBL entry should be treated as a starting hypothesis, not a fact. ## Finding the Function Section: A UniProt Database Tutorial Walkthrough ### Reading the Entry Page Start at uniprot.org and search by gene name, protein name, or accession. If you search by gene name, you will often get dozens of hits from different organisms. Filter to the organism you care about, and prefer the entry marked "reviewed" with a Swiss-Prot accession in the form of a letter plus five digits (P04637, Q9Y6K9, and so on). Click the accession to open the entry, then click Function in the left pane. On the Function tab, read the subsections top to bottom. The FUNCTION comment is a one-paragraph summary written for a knowledgeable biologist. Below it come the structured comments, then the cross-references to families, domains, pathways, and structures. Pay attention to the order in which the curators placed information. The most heavily evidenced and most central claims appear first. ### Reading the Flat File ![A branching flowchart showing how CC comment lines and FT feature lines in a UniProt flat file lead to comments, coordinates and evidence codes.](/article-images/illustrations/uniprot-function-annotation-explained-flat-file-reading.webp) The flat file is the authoritative machine-readable text. Retrieve it from the REST API with a single command: ``` curl https://rest.uniprot.org/uniprotkb/P04637.txt ``` In the output you will see the section structure directly. CC lines carry the comments, with `-!-` marking each individual comment item. FT lines carry features. A positional annotation looks like this: ``` FT ACT_SITE 179 FT /note="Proton acceptor" FT /evidence="ECO:0000269|PubMed:12345678" ``` The first token after FT is the feature key. ACT_SITE marks an active site residue, BINDING marks a binding site, SITE marks a residue of interest, DOMAIN marks a domain, REGION marks a region of interest, and MOTIF marks a short conserved motif. The number is the residue position in the displayed sequence. The note describes the role, and the evidence line gives an ECO code plus the source. Learning to read this block directly is the fastest way to extract positional annotations for any purpose, including mapping variants from genome coordinates onto protein residues [4]. ### Querying Through the REST API For more than one protein, use the search endpoint. A query that pulls the Function comment, EC number, and Gene Ontology terms for all reviewed human proteins looks like this: ``` curl "https://rest.uniprot.org/uniprotkb/search?query=reviewed:true+AND+organism_id:9606&fields=accession,id,protein_name,cc_function,ec,go_id&format=tsv&size=500" ``` The field names map one to one to what you see in the interface. cc_function returns the FUNCTION comment, cc_catalytic_activity returns catalytic activity comments, ft_act_site returns active sites with coordinates, ft_bind returns binding sites, keyword returns UniProt keywords, and go_id returns GO terms. Adding `&format=tsv` gives a tab-delimited table you can drop straight into R or pandas. This single pattern is enough to build most annotation workflows, and it is how the tools discussed later in this article feed UniProt data into their own pipelines. ## The Subsections of the Function Section, One by One ### The FUNCTION Comment The FUNCTION comment is the entry headline. It is written in the active voice, states the biological role, and names the process, the substrate, and often the regulatory context. A typical kinase FUNCTION comment reads as a sentence such as "Receptor for cytokines and growth factors that mediates [cell proliferation](/blog/guides/cell-proliferation) and differentiation through activation of downstream signaling pathways," followed by details of the ligand and pathway. Interpret it as a synthesis, not a proof. Curators compress dozens of papers into one paragraph, and they deliberately omit hedging. The evidence tags attached to the FUNCTION comment point back to the key papers. If the function is inferred only from a homolog, the tag will say so, and the sentence may name the homolog explicitly. ### Catalytic Activity, Cofactors, and Regulation For enzymes, the CATALYTIC ACTIVITY comment states the reaction in standardized form, for example "ATP + a [protein]-L-tyrosine = ADP + a [protein]-L-tyrosine phosphate," followed by the EC number. The COFACTOR comment lists required ions and small molecules with their positions in the sequence when they are ligands, for example a zinc coordinated by a set of cysteine residues. The ACTIVITY REGULATION comment describes activators, inhibitors, and phosphorylation states that switch the enzyme on or off. Read these three comments together. An EC number tells you the class of reaction, the catalytic activity line tells you the substrates, and the regulation comment tells you when the enzyme is actually active in the cell. Enzyme annotations of this kind are precise enough to support genome-scale comparisons, and studies have mined UniProt EC annotations across thousands of microbial genomes to correlate enzyme repertoire with physiological properties such as optimal growth temperature [5]. ### Subunit, Interaction, and Location The SUBUNIT comment describes the quaternary structure, for example "homodimer," "heterotrimer," or "monomer." The INTERACTION subsection (rendered under Interactions on the web page but logically part of functional knowledge) lists experimentally determined binding partners with the number of independent publications. The LOCATION comment states subcellular compartments, using the controlled vocabulary of the Gene Ontology cellular component branch plus UniProt's own subcellular location vocabulary. These three comments together answer the question of where the protein acts and with whom. A protein annotated as a nuclear phosphoprotein that homodimerizes and binds DNA is telling you a very different story from a secreted peptidase, and the Function section delivers that distinction before you ever look at a domain diagram. ### Pathway, Disease, and Tissue Specificity The PATHWAY comment describes membership in metabolic and signaling pathways, usually with cross-references to Reactome or KEGG. The DISEASE comment lists associations with human disease, each with a short description of the mechanism and the evidence. UniProtKB/Swiss-Prot provides curated knowledge on the functional impact and clinical significance of over 30,000 human protein-coding sequence variants, curated from peer-reviewed literature, and this variant knowledge is tightly integrated with the functional claims in the same entry [6]. The TISSUE SPECIFICITY comment, when present, describes where the protein is expressed at the protein level, which matters when you are deciding whether a protein-level observation in a given tissue is plausible. ### Post-Translational Modifications The PTM subsection lists phosphorylation sites, glycosylation sites, acetylation sites, ubiquitination sites, and more, each with coordinates and evidence. Because PTMs frequently gate activity, a phosphorylation site in an activation loop plus an ACTIVITY REGULATION comment describing that phosphorylation is often the mechanistic core of the whole entry. Genome-scale integrative studies routinely draw PTM characteristics from UniProt alongside specialized resources to build protein-variant-modification-disease networks [7]. ### Caution and Miscellaneous Notes Two comments deserve special attention because they can invalidate other annotations. CAUTION flags a claim that is disputed, erroneous, or under re-evaluation. MISCELLANEOUS holds information that does not fit elsewhere. If a CAUTION line exists in an entry you are about to rely on, resolve it before building anything downstream on the affected claim. ## Positional Annotations: Features That Carry Coordinates The Features table is the positional backbone of the Function section. The most important feature keys, in the order you will encounter them most often, are: | Feature key | What it marks | Typical note content | |--|--|--| | ACT_SITE | Active site residue | Catalytic role, such as proton acceptor | | BINDING | Binding site residue(s) | Ligand, such as ATP, substrate, or metal | | SITE | Residue of interest | Autophosphorylation, cleavage, or isomerization | | DOMAIN | Domain extent | Domain name and family | | REGION | Region of interest | Functional region not matching a named domain | | MOTIF | Short conserved motif | Consensus signature, such as an ATP-binding lysine | | ZN_FING | Zinc finger | Coordination residues | | MOD_RES | Modified residue | Phosphorylation, acetylation, or other PTM | | VAR_POS or variant | Variant position | Disease or benign amino acid change | Each feature has a start position, an optional end position (single-residue features have only a start), a note, and an evidence tag. Two attributes divide the table in two. Experimental features carry direct evidence. Computed features, produced by automatic annotation, are labeled with a rule source and displayed in the web interface in a different color from experimentally determined ones. A computed DOMAIN can still be extremely useful, because it usually comes from a well-tested signature model, but you should always know which color you are looking at. The coordinates are relative to the displayed sequence, which accounts for signal peptides and propeptides, so a feature position in the flat file may not match a genomic coordinate. UniProt provides positional annotations such as active sites, binding sites, and variants mapped to the [human genome](/blog/guides/human-genome) (GRCh38) as downloadable track hubs for UCSC and Ensembl genome browsers, which solves the coordinate problem for human work and allows computational comparison of protein annotations with clinical variant resources [4]. ProtVista renders the entire feature set graphically against the sequence, alongside experimental proteomics data and variation datasets, so you can see whether a variant of interest falls inside an active site or a domain at a glance [2]. This visual check should be part of every variant interpretation workflow, and it is the fastest way to build intuition for how features are distributed along a protein. ## Evidence Tags: The Most Important Part of the Section ![A three-question triage path that sorts a UniProt Function claim into experimental, similarity-based or automatic evidence and says what to do next.](/article-images/illustrations/uniprot-function-annotation-explained-evidence-triage.webp) Every Function statement and every feature carries at least one evidence tag, and the tags use the Evidence and Conclusion Ontology (ECO). Three families of codes cover nearly everything you will see: - Experimental evidence, coded ECO:0000269 in flat files, means the claim traces to a direct experimental result in a cited publication. This is the strongest tier for biological claims. - Sequence similarity evidence, coded ECO:0000250, means the claim was transferred from an experimentally characterized homolog. The tag usually names the source entry, so you can trace the original experiment. - Automatic assertion evidence, including ECO:0000256 for rule-based annotations and ECO:0000501 for signature matches, means a computational pipeline produced the claim without human review. In the web interface these tiers are color-coded, so a page can be scanned for gold tags (experimental) versus blue and gray tags (similarity and automatic). In the flat file you read the ECO code directly. The automatic tier is far more sophisticated than a simple BLAST hit. InterProScan, the annotation pipeline that drives large-scale [protein function](/blog/guides/protein-function) classification for UniProt, Ensembl, and MGnify, matches sequences against integrated signature databases such as Pfam, PROSITE, and PANTHER, and recent versions add modern deep-learning predictors alongside profile-based methods [8]. When a signature match plus a supporting rule fires, UniProt can propagate functions such as catalytic activity, GO terms, and domain architecture to thousands of related sequences at once. That is why an unreviewed TrEMBL entry can carry a rich-looking Function section with zero experimental work behind it. The evidence tags are the only way to tell. ## Gene Ontology, Keywords, and EC Numbers in the Function Context Three controlled vocabularies sit at the bottom of the Function section, and they are what most computational consumers actually parse. UniProt keywords are short controlled terms such as "Hydrolase," "Membrane," "Signal," and "Disease mutation." They are coarse but reliable, and they are the fastest filter for partitioning a proteome. Gene Ontology terms come in the three familiar branches. Molecular function terms describe activities, biological process terms describe objectives, and cellular component terms describe locations. Each GO term in the entry carries its own evidence tag, using GO evidence codes that interoperate with ECO, so an IMP tag (inferred from mutant phenotype) is experimentally grounded while an IEA tag (inferred from electronic annotation) is automatic. EC numbers classify enzymatic reactions hierarchically, and they are attached through the CATALYTIC ACTIVITY comment. Because EC numbers are standardized, they support genome-scale functional comparisons. EnzymeHunter, a deep-learning framework for fine-grained enzyme function prediction, uses the EC hierarchy to guide its loss function, and in a proteome-wide application to Thermus thermophilus it discovered novel catalytic functions, one of which was subsequently validated by an independent UniProt update [9]. This illustrates the feedback loop in modern annotation: UniProt's structured EC and GO terms serve both as training targets for predictors and as the venue where validated predictions eventually land. ## A Worked Example: Reading a Kinase Entry in Ten Minutes Here is how to read the Function section of a reviewed human protein tyrosine kinase entry in a deliberate order. Step one, read the FUNCTION comment. It should state the receptor or cytosolic kinase role, the ligand or activating condition, and the major downstream pathway. If the FUNCTION comment is vague, the rest of the entry probably is too. Step two, find the CATALYTIC ACTIVITY comment and confirm the reaction and EC number. For a tyrosine kinase the reaction is ATP plus a protein tyrosine residue yielding ADP and a phosphorylated tyrosine, with an EC number in the 2.7.10 class. If the entry is a pseudokinase, the catalytic activity comment will be absent or explicitly marked as inactive, and the CAUTION comment will often explain why. Step three, locate the ACT_SITE and BINDING features. The active site lysine in the ATP-binding lobe and the catalytic residues in the activation loop should carry experimental or rule-based evidence. Open ProtVista and confirm that these positions fall inside the annotated [protein kinase](/knowledge/molecular-biology/protein-kinase) domain rather than outside it [2]. Step four, read ACTIVITY REGULATION and PTM together. A phosphorylation site in the activation loop paired with a regulation comment describing autophosphorylation tells you the switching mechanism. A kinase lacking both is either constitutively active or incompletely characterized, and the evidence tags will tell you which. Step five, check SUBUNIT, LOCATION, and PATHWAY. A [receptor tyrosine kinase](/knowledge/molecular-biology/receptor-tyrosine-kinase) should show membrane location, dimerization, and membership in a signaling pathway. The DISEASE comment should list activating mutations, and clicking through to the variant features shows where those mutations sit relative to the domains and active sites. Step six, scan every evidence tag you plan to rely on. If the central claims are experimental, you can build on the entry. If the central claims are similarity-based, trace the tag back to the source entry and read that one. If the claims are purely automatic, treat the entry as a hypothesis and confirm at least the one fact your analysis actually needs. ## How the Function Section Feeds Real Workflows ### Variant Interpretation Variant interpretation is the highest-stakes consumer of the Function section. UniProt has mapped human sequences and positional annotations, including active sites, binding sites, and variants, to the human genome and released public track hubs, demonstrating that most UniProt disease-associated variants that colocalize with ClinVar SNPs map to variants classified as pathogenic [4]. The curatorial workflow behind this has been standardized and documented, and re-curation of UniProtKB/Swiss-Prot variants according to American College of Medical Genetics and Genomics (ACMG) guidelines using ClinGen tools increased concordance with ClinVar, mainly by reclassifying supposedly pathogenic variants as benign on the basis of population frequency evidence [6]. Practically, this means that when you see a DISEASE comment in a Swiss-Prot entry, you are reading a claim that has been through a documented curation pipeline and is being actively harmonized with ClinVar. Tools such as Genome Nexus aggregate variant information from more than a dozen resources, including UniProt [protein sequence](/blog/guides/protein-sequence) annotation alongside Pfam and dbPTM domain and modification data, so that a clinician can see variant effect predictions, protein features, and population frequencies in one view [10]. Language-model frameworks for variant evidence extraction also consume UniProt domain annotations as structured inputs to ACMG/AMP rule engines [11]. In every one of these pipelines, the Function section is the layer that converts a raw variant coordinate into a mechanistic hypothesis. ### Proteomics and PTM-Centric Analysis Proteomics pipelines consume UniProt functional annotations to turn lists of identified peptides into biological statements. ProteoSushi, a tool for annotating redox and other modification-specific proteomics datasets, assigns each modified peptide to shared proteins and annotates each modified site with current UniProt information, including known functional roles of the site and the domains it falls within [12]. Lipid raft proteomics databases have similarly re-mapped all entries to UniProt identifiers and implemented annotation based on experimental evidence level to help researchers distinguish bona fide raft proteins from contaminants [13]. When you interpret a proteomics hit list, the evidence tier of the UniProt Function section is what lets you distinguish a well-characterized protein from one whose raft association appeared once in one study. ### Metagenomics and Transcriptomics Annotation Environmental and transcriptomic datasets produce enormous numbers of proteins with no experimental history, and UniProt annotations are the standard bridge to interpretation. The annoreport tool aggregates gene annotation across metagenome-assembled genomes and performs functional gene clustering via UniProt annotation, reporting category distributions such as Transport and Membrane, Nucleotide Binding, and DNA Metabolism across thousands of coding sequences [14]. De novo transcriptomic assembly studies in non-model organisms likewise report the fraction of transcripts functionally annotated against UniProt, GO, and KEGG as a headline result, because the fraction of annotated transcripts directly measures how much biological interpretation is possible [15]. ### Function Prediction and Machine Learning A growing family of tools uses UniProt as both a data source and a benchmark. ProtPen combines sequence-based annotation from eggNOG-mapper with structure-based searches of AlphaFold-predicted models, then merges and enriches the results with UniProt metadata, achieving annotation accuracy above 90 percent on a benchmarked Pseudomonas aeruginosa dataset [16]. ViralMultiNet, a structure-aware framework for viral protein function prediction, derives functional semantic embeddings from UniProt annotations and integrates them with k-mer sequence encodings for wastewater surveillance applications [17]. Most strikingly, Affinage applies a large language model pipeline to the primary literature and generates mechanistic annotations for 19,293 human protein-coding genes, providing mechanism for thousands of genes whose UniProt Function section is empty or a stub [18]. That last point matters for readers: an empty or thin Function section does not mean the protein has no known function, it means UniProt has not yet captured one, and the literature may be ahead of the database. ### Structural Mapping Structure is the final layer of functional interpretation. The PDBe-SIFTS framework provides residue-level mappings between UniProt Knowledgebase sequences and [Protein Data Bank](/blog/guides/protein-data-bank) structures, now as a fully open-source pipeline that recovered curated mappings at top rank in 93.1 percent of benchmarked cases across the complete PDB archive [19]. When you read a UniProt entry and see a cross-reference to a PDB structure, that link was established through mapping of exactly this kind. Structural and functional assessment against UniProt and the Protein Data Bank is also standard practice in virology annotation workflows, such as systematic reviews that map viral variants onto protein regions to judge their relevance [20]. ## Summary Table | Item | What it tells you | Where to check it | Typical strongest evidence | |--|--|--|--| | FUNCTION comment | One-paragraph biological role | First CC block, web Function tab | Experimental, cited by PMID | | CATALYTIC ACTIVITY | Reaction and EC number | CC block, EC field | Experimental or inferred by similarity | | COFACTOR | Required ions and cofactors | CC block | Experimental | | ACTIVITY REGULATION | Switches, inhibitors, activators | CC block | Experimental | | SUBUNIT and LOCATION | Quaternary structure and compartment | CC block | Experimental or inferred by similarity | | PATHWAY and DISEASE | Pathway membership and clinical relevance | CC block, cross-references | Experimental, curated | | Features (ACT_SITE, BINDING, DOMAIN) | Residue coordinates of functional elements | FT lines, ProtVista view | Experimental versus computed, color-coded | | GO terms and keywords | Controlled vocabulary descriptors | GO and keyword fields | Mixed, check each tag | | Evidence tags (ECO) | Basis of every claim above | Tag hover in web, ECO code in flat file | ECO:0000269 is the gold standard | ## Common Mistakes and Limitations The most common and consequential mistake is reading a TrEMBL entry as if it were curated. Automatic annotation pipelines produce coherent, plausible, well-formatted Function sections, and a casual reader cannot tell them from curated text without checking the evidence tags. Always confirm the reviewed status and then check at least the tags on the claims your analysis depends on. A second mistake is ignoring the difference between experimental and computed features. Computed domains and rule-based active site predictions are frequently correct, but a computed ACT_SITE is a prediction about a specific residue, and designing a mutation experiment or interpreting a variant on the basis of an unverified prediction is a different risk level than doing so on an experimentally confirmed site. A third mistake is treating a missing Function section as evidence of no function. Thousands of human genes have empty or stub-level UniProt function text despite published mechanistic literature, and the dark proteome, particularly proteins from uncultured microorganisms and phylogenetically underrepresented lineages, is becoming less visible as UniProt prunes redundant records [3] [18]. Absence of annotation is a statement about database state, not about biology. A fourth mistake is assuming all UniProt entries for the same protein agree. Entries for orthologs in different species carry different evidence levels, and an unreviewed human fragment can carry inherited annotations that a reviewed entry from another species properly grounds. When in doubt, work from the reviewed entry for your organism of interest and trace similarity tags back to the experimentally characterized source. A fifth mistake concerns coordinate systems. Feature positions refer to the displayed precursor sequence, including signal peptides and transit peptides, so a feature position may not equal a mature protein position or a genomic coordinate. For human work, use the genome-mapped positional annotations and track hubs rather than doing coordinate arithmetic by hand [4]. Finally, the Function section describes the canonical sequence, and it does not by itself resolve isoform-specific function, tissue-specific PTM patterns, or context-dependent interactions. Individual experimental or clinical questions need targeted literature and, where relevant, domain expertise, and the entry is a starting map rather than a final authority. ## Quick Review - The Function section combines a plain-language FUNCTION comment with structured comments (catalytic activity, cofactors, regulation, subunit, location, pathway, disease, PTM) and positional features (active sites, binding sites, domains). - Reviewed Swiss-Prot entries are curated by experts, unreviewed TrEMBL entries are computationally generated, and the difference determines how much you can trust the text. - Every claim carries an evidence tag, and the ECO code separates experimental, similarity-based, and automatic tiers. - In the flat file, CC lines hold comments and FT lines hold features with residue coordinates, notes, and evidence. - The REST API exposes the same content through field names such as cc_function, ft_act_site, ec, and go_id, which makes bulk extraction straightforward. - Computed features are visually distinct from experimental features, and computed does not mean wrong, it means unverified at that residue. - An empty Function section reflects the state of the database, not the state of knowledge, since only about 2 percent of UniProt entries are experimentally validated [1]. ## Frequently Asked Questions ### What does the Function section of a UniProt entry actually describe? It describes what a protein does, in two layers. The first layer is prose: a FUNCTION comment and related comments on catalytic activity, cofactors, regulation, subunit structure, location, pathway, disease, and post-translational modification. The second layer is positional: features such as active sites, binding sites, and domains, each with residue coordinates and an evidence tag stating how the claim was established. ### How do I tell experimental annotation from prediction in UniProt? Check the evidence tags. In the web interface, hover over the small colored tags to see the source, with experimental claims distinguished from similarity-based and automatic claims by color. In the flat file, read the ECO code, where ECO:0000269 marks experimental evidence, ECO:0000250 marks inference from a homolog, and codes such as ECO:0000256 and ECO:0000501 mark automatic annotation. ### Why is the Function section empty or very short for my protein? Because UniProt has not captured published knowledge for that entry yet, or because the protein genuinely lacks characterization. Only about 2 percent of the over 203 million UniProt entries have experimental validation, so thin Function sections are the norm for unreviewed entries [1]. For human proteins, recent literature-mining pipelines show that many genes with empty or stub-level UniProt functions do have mechanistic information in the primary literature [18]. ### What is the difference between the Function section and the Features section? The Function section is conceptual, describing roles, reactions, and associations. The Features table is positional, assigning residue coordinates to those concepts, such as which residue is the catalytic base or which stretch of sequence forms a domain. The two are linked, and ProtVista draws the positional layer along the sequence so you can see whether a variant or modification overlaps a functional feature [2]. ### Can I cite a UniProt entry directly in a paper? Yes, and the correct practice is to cite the UniProt resource along with the specific accession and release version, and to cite the primary papers behind any experimental claim you rely on. The evidence tags inside the entry give you those PubMed identifiers. For claims you borrow without verifying the primary literature, state the evidence level in your methods, since a similarity-based or automatic claim is not equivalent to an experimental one. ### How do I download Function annotations for many proteins at once? Use the UniProt REST API search endpoint with the fields you need. A query such as `https://rest.uniprot.org/uniprotkb/search?query=reviewed:true+AND+organism_id:9606&fields=accession,cc_function,ec,go_id&format=tsv` returns a tab-delimited table of accession, function text, EC numbers, and GO terms for all reviewed human proteins. Larger jobs can use the UniProt FTP site or the streaming endpoint, and the same field names work regardless of how many accessions you retrieve. ## Sources 1. [Integrating Protein Language Models with Multimodal Embeddings to Accelerate Function Prediction of Uncharacterized Proteins.](https://pubmed.ncbi.nlm.nih.gov/42123473/) 2. [ProtVista: visualization of protein sequence annotations.](https://pubmed.ncbi.nlm.nih.gov/28334231/) 3. [UniProt data pruning and the growing invisibility of the dark proteome.](https://pubmed.ncbi.nlm.nih.gov/42889534/) 4. [UniProt genomic mapping for deciphering functional effects of missense variants.](https://pubmed.ncbi.nlm.nih.gov/30840782/) 5. [Correlating enzyme annotations with a large set of microbial growth temperatures reveals metabolic adaptations to growth at diverse temperatures.](https://pubmed.ncbi.nlm.nih.gov/30400856/) 6. [An enhanced workflow for variant interpretation in UniProtKB/Swiss-Prot improves consistency and reuse in ClinVar.](https://pubmed.ncbi.nlm.nih.gov/30937429/) 7. [Systems biology-driven analysis of obesity-associated genes with transcript mapping and post-translational modification annotations.](https://pubmed.ncbi.nlm.nih.gov/42172902/) 8. [InterProScan 6: a modern large-scale protein function annotation pipeline.](https://pubmed.ncbi.nlm.nih.gov/42222668/) 9. [EnzymeHunter: Achieving fine-grained enzyme function prediction with a hierarchically aware contrastive learning framework.](https://pubmed.ncbi.nlm.nih.gov/42453691/) 10. [Genome Nexus: A Comprehensive Resource for the Annotation and Interpretation of Genomic Variants in Cancer.](https://pubmed.ncbi.nlm.nih.gov/35148171/) 11. [OtoVCE: a mechanism-aware language-model evidence layer for hereditary hearing-loss variant interpretation.](https://pubmed.ncbi.nlm.nih.gov/42644181/) 12. [Automating Assignment, Quantitation, and Biological Annotation of Redox Proteomics Datasets with ProteoSushi.](https://pubmed.ncbi.nlm.nih.gov/35604553/) 13. [RaftProt V2: understanding membrane microdomain function through lipid raft proteomes.](https://pubmed.ncbi.nlm.nih.gov/30329070/) 14. [annoreport: an interactive tool for metagenome annotation.](https://pubmed.ncbi.nlm.nih.gov/42763764/) 15. [Transcriptomic response of the wild plant Mesembryanthemum forsskalii Hochst. ex Boiss to salinity stress: de novo assembly, functional annotation, and transcription factor profiling.](https://pubmed.ncbi.nlm.nih.gov/42376256/) 16. [ProtPen Combines Sequence- and Structure-based Approaches to Facilitate Protein Function Predictions on a Proteome-wide Scale.](https://pubmed.ncbi.nlm.nih.gov/42695892/) 17. [ViralMultiNet: A structure-aware multimodal framework for viral protein function prediction in wastewater surveillance.](https://pubmed.ncbi.nlm.nih.gov/42234710/) 18. [Affinage: Genome-Scale Mechanistic Gene Annotation from the Published Literature.](https://pubmed.ncbi.nlm.nih.gov/42428021/) 19. [PDBe-SIFTS: An open-source tool for Structure Integration with Function, Taxonomy, and Sequences, featuring improved alignment, scoring scheme, and accelerated search.](https://pubmed.ncbi.nlm.nih.gov/42885203/) 20. [Human papillomavirus 16 genomic variation in head and neck cancer: a systematic review with structural and functional annotation.](https://pubmed.ncbi.nlm.nih.gov/43003906/)