# UniProt Annotations for Experimental Design: A Guide Most failed cloning, expression, antibody, and mutagenesis experiments trace back to a sequence-level decision made before the first pipette was picked up: the construct missed a [signal peptide](/knowledge/molecular-biology/signal-peptide), the antibody peptide fell inside a transmembrane helix, the point mutation hit a post-translational modification site, or the "novel" protein turned out to be an unreviewed automatic prediction. UniProt annotations exist to prevent exactly these mistakes, and this guide shows you how to use them deliberately. By the end of this article you will be able to find the correct UniProt entry for your protein, read an entry the way an experimental planner reads it, pull sequences and annotation tables with the UniProt REST API, map gene symbols and foreign identifiers to stable accessions, download a whole proteome for batch work, and translate specific annotation lines into concrete design choices such as construct boundaries, peptide targets, and mutation lists. What you need on hand: - A web browser for interactive work at uniprot.org - Python 3.8 or newer with the `requests` library installed, or `curl` on the command line - A text editor or notebook for the small scripts shown here - Optional but useful: a local BLAST installation if you work with unsequenced or divergent organisms ## What UniProt Is and Why Its Annotations Matter UniProt is the primary provider of [protein sequence](/blog/guides/protein-sequence) and annotation data for the life sciences. The UniProt consortium formed in 2002 from groups at the Swiss Institute of Bioinformatics, the [European Bioinformatics Institute](/knowledge/bioinformatics/the-european-bioinformatics-institute-embl-ebi), and the Protein Information Resource at Georgetown University, and uniprot.org has served as the central entry point to its data since 2008 [1]. The database grew past 80 million sequences as genome projects flooded it, which forced two changes that matter directly to experimentalists: the accession format expanded from 6 to 10 characters, and a new proteome identifier was introduced so that a specific assembly of a species and strain can be tracked unambiguously [2]. Two facts about how UniProt is organized shape everything else in this guide. First, UniProtKB splits into two tiers. **UniProtKB/Swiss-Prot** entries are reviewed by expert curators who integrate literature, experimental data, and computational predictions. **UniProtKB/TrEMBL** entries carry computationally generated annotations awaiting full review. For experimental design, reviewed entries are your default source of truth. Unreviewed entries are useful for coverage of poorly studied organisms, but their annotations carry weaker evidence, a distinction explained below. Second, every entry carries an **annotation score**, a value from 1 to 5 that summarizes how much is known about a protein. Proteins with high scores are well characterized and richly annotated, which makes them strong candidates for comparative analysis, control proteins, or references when you characterize a homolog [2]. A score of 1 or 2 tells you up front that you are working near the edge of known biology and should expect to validate annotations yourself. A quick mental model for the rest of this guide: a UniProt entry is not just a sequence. It is a structured claim sheet about that sequence, where each claim has an attached evidence type. Experimental design is the process of converting those claims into decisions about DNA, reagents, and assays. ## The Annotation Categories That Drive Experimental Decisions UniProt entries store information as feature keys, and each feature key maps naturally onto a class of experimental decision. The table below lists the features you will reach for most often, the REST API field name used to retrieve them in bulk, and the design question each one answers. | Feature or annotation | REST field name | Design question it answers | |--|--|--| | Signal peptide | `ft_signal` | Where does the mature protein start for secreted constructs? | | Transmembrane region | `ft_transmem` | Can this be expressed soluble, or does it need detergent, nanodiscs, or a cell-based assay? | | Topological domain | `ft_topo_dom` | Which loops or tails face the cytoplasm versus the extracellular space for epitope or sensor design? | | Domain (InterPro, Pfam, PROSITE) | `ft_domain`, `xref_interpro` | What are the natural module boundaries for truncation constructs? | | Active site, binding site | `ft_act_site`, `ft_binding` | Which residues must stay intact for catalytic or ligand-binding assays? | | Modified residue, glycosylation | `ft_mod_res` | Which sites should be preserved, mutated, or expressed in a eukaryotic system? | | Variant (natural variants) | `ft_variant` | Do disease variants or isoform differences conflict with my mutation plan? | | Alternative products (isoforms) | `ft_var_seq` | Which isoform matches my tissue, my antibody, or my cDNA clone? | | Subcellular location | `cc_subcellular_location` | Where should my tagged construct localize, and which compartment lysis do I need? | | Tissue specificity | `cc_tissue_specificity` | Is the target expressed in the cells or tissue my assay assumes? | | Interaction partners | `cc_interaction` | Do I need a co-expression partner or a complex assay rather than the protein alone? | | PDB and AlphaFold cross-references | `xref_pdb`, `xref_alphafold` | Is there a structure to guide boundary or mutation choices? | | Sequence caution | `cc_seq_caution` | Is the underlying sequence itself suspect? | Two habits will pay off more than any single feature. First, always read the evidence attached to a feature. UniProt marks each annotation with an evidence type, and manual curation backed by experiment is a materially different claim from automatic annotation propagated by similarity. Second, always check the entry history and version. Entries are updated continuously, and an annotation set you downloaded six months ago may have changed. ## How to Use UniProt, Step by Step This walkthrough follows a realistic project: you need to design a soluble expression construct, an antibody peptide, and a small mutagenesis panel for the human epidermal growth factor receptor, EGFR. The commands generalize to any target. ### Step 1. Find and verify the correct entry Search uniprot.org for the gene symbol plus the organism, for example `EGFR AND human`. The result list shows one reviewed canonical entry, **P00533 (EGFR_HUMAN)**, alongside any TrEMBL entries from redundant genome submissions. Verified criteria for the right entry: - The status is **Reviewed (Swiss-Prot)** - The organism matches your experiment exactly, down to strain if you work in a non-standard isolate - The gene name matches your source material, since paralogs such as ERBB2 and ERBB3 sit one click away - The sequence length matches expectation for the isoform you care about A common trap at this stage is confusing the **entry name** (EGFR_HUMAN) with the **accession** (P00533). Entry names are stable within releases but the accession is the permanent identifier you should cite, order primers against, and record in lab notebooks. Where isoforms exist, the canonical sequence shown by default may not be the one your cDNA encodes. Open the "Sequences" section, compare each isoform against your clone by alignment, and record the exact isoform accession before ordering anything. ### Step 2. Read the entry as an experimental plan ![Five-step order for reading a UniProt entry: processing, topology, domains and sites, modifications, then variants.](/article-images/illustrations/uniprot-annotations-experimental-design-entry-read-order.webp) Open P00533 and read the "Features" table in a fixed order that mirrors how designs fail: 1. **Processing.** The signal peptide spans residues 1 to 24 for human EGFR, so a soluble ectodomain construct should start near residue 25, not at residue 1. The mature chain boundary and any propeptide or transit peptide lines tell you where the biological sequence actually begins. 2. **Topology.** A single transmembrane helix spans roughly residues 646 to 668, splitting the protein into a large extracellular ligand-binding region and an intracellular kinase region. Any construct spanning the membrane requires detergents or a membrane expression strategy. 3. **Domains and sites.** The [protein kinase](/knowledge/molecular-biology/protein-kinase) domain sits around residues 712 to 979, with active-site residues annotated inside it. Active-site residues are hard constraints: mutations there abolish catalytic activity, which is either your goal (kinase-dead controls) or your failure mode. 4. **Post-translational modifications.** Glycosylation and phosphorylation sites are annotated with evidence. A glycosylation site inside a bacterial expression construct is a warning that the protein may misfold or aggregate in E. coli. 5. **Variants.** Natural variant annotations flag residues where point mutations are already known to alter function, useful both as positive controls and as regions to avoid accidentally hitting. This ordered read-through takes ten minutes per protein and routinely changes construct plans. Domain databases such as InterPro and Pfam supply the domain lines, and external resources build directly on them. OpenAntigens, for example, integrates UniProt topology, InterPro and Pfam domains, AlphaFold confidence values, and PDB precedent to generate construct-design reports for thousands of human secreted and membrane proteins, reflecting exactly this design logic at scale [3]. ### Step 3. Pull sequences and annotations with the REST API Interactive reading works for one protein. For panels, screens, or meta-analyses, use the REST API at `rest.uniprot.org`. Every UniProt website view has a programmatic equivalent. Retrieve the FASTA sequence: ```bash curl -s "https://rest.uniprot.org/uniprotkb/P00533.fasta" ``` Expected output (abbreviated): ``` >sp|P00533|EGFR_HUMAN Epidermal growth factor receptor OS=Homo sapiens OX=9606 GN=EGFR PE=1 SV=1 MRPSGTAGSGNALLALNLLP... ``` The FASTA header itself is informative: `sp` marks reviewed status, `P00533` is the accession, `EGFR_HUMAN` the entry name, and `SV=1` the sequence version. If the sequence version changes after you download, the residues changed, which matters for reproducing published constructs. Pull a structured annotation table as tab-separated values: ```bash curl -s "https://rest.uniprot.org/uniprotkb/search" \ -d "query=(gene:EGFR) AND (organism_id:9606) AND (reviewed:true)" \ -d "fields=accession,id,protein_name,length,annotation_score,ft_signal,ft_transmem,ft_domain,xref_alphafold" \ -d "format=tsv" ``` Expected output (abbreviated, column headers included by the API): ``` Accession Entry Protein_Name Length Annotation_Score Signal_peptide Transmembrane Domain AlphaFold P00533 EGFR_HUMAN Epidermal growth factor receptor 1210 5 SIGNAL 1..24 TRANSMEM 646..668 PROTEIN_KINASE ... AF-P00533-F1 ``` The same query in Python, filtering features to a compact design summary: ```python import requests response = requests.get("https://rest.uniprot.org/uniprotkb/P00533.json") data = response.json() design_features = {"Signal", "Transmembrane", "Domain", "Active site", "Glycosylation"} for feature in data["features"]: if feature["type"] in design_features: start = feature["location"]["start"]["value"] end = feature["location"]["end"]["value"] description = feature.get("description", "") print(f"{feature['type']:16s} {start:>5} - {end:<5} {description}") ``` Expected output (abbreviated): ``` Signal 1 - 24 Signal peptide Transmembrane 646 - 668 Helical Domain 712 - 979 Protein kinase Active site 813 - 813 PROTON ACCEPTOR Glycosylation 32 - 32 N-linked (GlcNAc...) asparagine ``` Always verify an API result against the web page at least once per project. If a field comes back empty when you expected content, the most common cause is a field name typo or an organism filter that excluded your entry, not missing data. ### Step 4. Map foreign identifiers to UniProt accessions Datasets from RNA-seq, proteomics, and chemical biology arrive keyed to gene symbols, RefSeq, Ensembl, or PDB identifiers, none of which are stable across pipeline changes. The UniProt ID mapping service converts them in one batch, a capability the consortium has provided as a core website tool since its first unified site [1]. ```bash curl -s -X POST "https://rest.uniprot.org/idmapping/run" \ -d "from=Gene_Name&to=UniProtKB&ids=EGFR,TP53,BRCA1" ``` Expected output: ``` {"jobId":"9e8b3f2c4a1d5e6f"} ``` Poll the status endpoint with that identifier: ```bash curl -s "https://rest.uniprot.org/idmapping/status/9e8b3f2c4a1d5e6f" ``` When the job reports completed, fetch the results and inspect them. One gene symbol can map to several accessions, including isoform-specific and TrEMBL entries. Resolve ambiguities by filtering for `reviewed:true` and the correct organism, and record the mapping date, because cross-references shift between releases. Chemical biology projects face the same standardization problem on the ligand side, which is why dedicated platforms now provide expert, controlled annotation of both targets and compounds to make docking predictions and experimental binding data comparable [4]. ### Step 5. Batch-download a proteome when working at scale When you need every protein from an organism, query by proteome identifier rather than downloading entry by entry. The human reference proteome is UP000005640: ```bash curl -s "https://rest.uniprot.org/uniprotkb/stream" \ -d "query=(proteome:UP000005640) AND (reviewed:true)" \ -d "format=fasta" \ -d "compressed=true" \ -o human_reviewed.fasta.gz ``` Expected result: a gzipped FASTA containing the reviewed human proteome, suitable for BLAST database construction, peptide synthesis planning, or Mascot and MaxQuant search preparation. To pull annotations alongside sequences, switch `format=tsv` and append the field list from Step 3. The proteome identifier system exists precisely so that work against a specific species and strain assembly stays traceable as genomes get re-sequenced [2]. Check the result before building anything downstream: count sequences with `grep -c ">"`, confirm the expected magnitude for your organism's reviewed proteome, and spot-check two or three entries against the website. ### Step 6. Convert annotations into concrete experimental decisions The final step is mechanical, and writing it out prevents drift between plan and bench work. For EGFR, the read-through in Step 2 converts directly into: - **Soluble ectodomain construct:** clone residues 25 through roughly 645, stopping before the transmembrane helix, and express in a system that performs N-linked glycosylation, since annotated glycosylation sites fall inside the ectodomain. - **Antibody peptide:** choose a 12 to 18 residue peptide from an annotated extracellular loop or tail region with no annotated modification site and no overlap with conserved kinase-domain residues, which reduces cross-reactivity risk against ERBB-family paralogs. - **Kinase-dead control:** mutate the annotated catalytic aspartate inside the protein kinase domain, and confirm in the natural variant table that your chosen substitution is not itself a known activating mutation. - **Assay interpretation:** if the subcellular location annotation lists both membrane and endosomal compartments, plan a fractionation or imaging step rather than assuming a single pool. Large-scale resources formalize this conversion. PrankWeb accepts a UniProt identifier directly and predicts ligand-binding sites, falling back to AlphaFold models when no experimental structure exists, which lets pocket-directed mutagenesis start from an accession alone [5]. AlphaSync maintains millions of structural models synchronized to current UniProt entries and layers on residue-level annotations such as solvent accessibility and disorder, covering proteins whose models in other resources are out of date relative to the latest sequence [6]. Machine-learning resources for active-site classification build their training sets from UniProt annotations paired with AlphaFold2 structures, which means the annotations you read are the same substrate those tools learn from [7]. ## UniProt Annotations in Real Workflows, From Vaccines to Receptor Engineering Seeing how published projects consume UniProt annotations makes the design logic concrete. **Reverse vaccinology pipelines** source target protein sequences in FASTA format from UniProt and proteome databases, remove redundant and near-identical sequences with clustering tools such as CD-Hit, and then run epitope prediction for B-cell, cytotoxic T-cell, and helper T-cell responses before filtering candidates for antigenicity, toxicity, and allergenicity [8]. The UniProt contribution at the front of that pipeline is decisive: correct accession, correct strain proteome, and correct mature-chain boundaries, because a signal peptide left inside a vaccine construct changes the antigen entirely. **Construct design for difficult protein classes** is being standardized against UniProt topology data. OpenAntigens produces construct-design reports for thousands of human cell-surface and secreted proteins by reconciling UniProt topology, predicted and experimental structures, domain annotations, orthologs, and paralogs before any DNA is ordered, motivated by the observation that choosing which region of a protein to express has remained poorly standardized across antibody discovery, structural biology, and computational binder design [3]. **Receptor engineering** leans on the same annotations. GPCRchimeraDB catalogs over two hundred chimeric G protein-coupled receptors and links them to natural class A GPCRs, describing every entry at the sequence, structural, and biophysical level using annotations from resources including UniProt, which lets designers identify where the intracellular limits of transmembrane regions fall before merging receptors [9]. **Comparing computational and experimental binding data** requires clean target annotation as much as clean chemistry. The dockNmine portal was built because experimental binding data lack transferability and molecule names lack standardization, and it provides expert, authenticated annotation of both ligands and targets so docking predictions can be scored against real measurements [4]. If you dock against a protein whose UniProt entry disagrees with the PDB construct you actually modeled, the discrepancy usually lives in the annotations you skipped. ## UniProt Annotations Experimental Design Tips That Save Time These tips compress the recurring habits of people who use UniProt well. - **Filter to reviewed entries first.** Search with `reviewed:true` in the query bar or API. Expand to TrEMBL only when your organism has little reviewed coverage, and then treat the annotations as predictions. - **Sort candidates by annotation score.** When picking control proteins or homolog references, high annotation scores identify the best-characterized proteins and make comparative analysis more informative [2]. - **Record accession plus sequence version.** `P00533` with its sequence version pins the exact residues you designed against. - **Cross-check topology against structures.** UniProt topology lines and experimental structures agree overwhelmingly, but glycosylation, disorder, and cleavage sites are exactly where predictions and reality diverge, and AlphaFold confidence values flag those regions independently [6]. - **Use the cross-references outward.** The InterPro, Pfam, PDB, AlphaFold, and Expression Atlas links on each entry are curated jump points. Following them costs seconds and often settles a boundary dispute. - **Automate retrieval for anything you will do twice.** The REST API returns stable formats, and a saved script with your query and field list doubles as documentation of exactly which annotations informed the design. - **Download annotations and sequences from the same query.** Mixing a FASTA from one release with an annotation table from another creates silent off-by-a-few-residues errors in feature coordinates. ## Common Mistakes and Limitations ![Two-question triage for deciding whether a reviewed or unreviewed UniProt entry can support design work.](/article-images/illustrations/uniprot-annotations-experimental-design-reviewed-or-trembl.webp) **Treating unreviewed annotations as established fact.** TrEMBL entries carry automatically generated annotations transferred by similarity from other proteins. For a protein in a poorly sequenced genome, the "active site" line may be pure inference. Read the evidence tag on every annotation you act on, and when evidence is by similarity only, plan an assay that validates the assumption. **Designing against the wrong isoform.** [Alternative splicing](/blog/guides/alternative-splicing) means the canonical sequence is one choice among several. Antibodies raised against a peptide in a region absent from your expressed isoform will fail silently. Compare isoform sequences in the entry's "Sequences" section before ordering peptides or primers, and note that isoform coverage in structure databases can lag the current sequence set, which is a gap resources like AlphaSync exist to close [6]. **Ignoring sequence caution and version history.** Some entries carry explicit sequence caution notes, and sequences themselves get revised. A construct designed against an old sequence version may differ from the current one. Check the version you downloaded and re-check before resubmitting a synthesis order months later. **Copying mature-chain boundaries from memory.** Signal peptide and propeptide lengths vary between homologs, and memorized offsets from one protein get applied to another. Read the processing features for the specific entry every time. **Confusing prediction with measurement in structure choices.** AlphaFold-derived resources provide predicted structures of remarkable breadth, with over two million models synchronized to UniProt in some resources [6], but prediction confidence varies by region. Where a docking or binding prediction drives a wet-lab decision, tools that combine model data with experimental evidence and evolutionary conservation, such as ligand-binding site predictors that estimate conservation from curated sequence databases [5], give a firmer footing than a single model alone. **Letting identifier drift corrupt your dataset.** Gene symbols are ambiguous across species and change over time, and merged or demerged accessions accumulate. Route every external identifier list through the UniProt ID mapping service, resolve one-to-many mappings explicitly, and archive the mapping output with your notebooks. **Expecting annotations to replace pilot experiments.** UniProt annotations concentrate what is known, including what is known only by inference. Well-annotated high-score entries support confident design, low-score entries support hypotheses. Build a small validation step, such as a solubility test expression or a single control antibody, into plans that rest on thin annotations. **Forgetting that annotations reflect the literature.** Annotation completeness varies by protein family, organism, and research fashion. A missing interaction line does not prove an interaction does not exist. When a critical annotation is absent, the entry tells you the state of public knowledge, not the state of nature, and a targeted literature search or pilot experiment fills the gap better than assumption. ## Frequently Asked Questions ### What is the difference between reviewed and unreviewed UniProt entries? Reviewed entries (Swiss-Prot) are curated by experts with literature-backed, evidence-tagged annotations. Unreviewed entries (TrEMBL) carry computationally predicted annotations and should be treated as starting points for verification. ### How do I cite UniProt in a paper? Cite the accession, the entry, and the UniProt release or database publication. The UniProt Consortium publishes periodic database papers in [Nucleic Acids](/blog/guides/nucleic-acids) Research that serve as standard citations [2]. ### Is the annotation score a measure of accuracy? No. The annotation score reflects the relative amount of knowledge known about a protein, indicating how well characterized it is, not whether each annotation is correct [2]. ### How do I choose the right isoform for my experiment? Compare each isoform in the entry's "Sequences" section against your clone or tissue evidence by alignment, confirm which exons your source material contains, and record the isoform-specific accession. ### How do I convert a list of gene names to UniProt accessions? Use the UniProt ID mapping service, posting your identifiers with a source type such as Gene_Name and a target of UniProtKB, then resolve any one-to-many results by filtering for reviewed entries and your organism. ### Can I download annotations for an entire proteome at once? Yes. Query the REST API stream endpoint by proteome identifier, such as UP000005640 for the human reference proteome, and select FASTA or TSV output, optionally compressed [2]. ### How do UniProt annotations connect to predicted structures? Entries cross-reference AlphaFold models, and dedicated resources keep structural predictions synchronized with the current UniProt sequence set, adding residue-level annotations such as disorder and solvent accessibility [6]. ### Are UniProt annotations reliable enough for primer and construct design? Reviewed annotations backed by experimental evidence are reliable enough to design against, while predictions should be spot-checked against structures, conservation, or a small pilot experiment before committing to synthesis orders. ## Sources 1. [Infrastructure for the life sciences: design and implementation of the UniProt website.](https://pubmed.ncbi.nlm.nih.gov/19426475/) 2. [UniProt: a hub for protein information.](https://pubmed.ncbi.nlm.nih.gov/25348405/) 3. [OpenAntigens: a structure-aware database for antigen construct design across the human cell-surface and secreted proteome.](https://pubmed.ncbi.nlm.nih.gov/42952403/) 4. [DockNmine, a Web Portal to Assemble and Analyse Virtual and Experimental Interaction Data.](https://pubmed.ncbi.nlm.nih.gov/31614716/) 5. [PrankWeb 3: accelerated ligand-binding site predictions for experimental and modelled protein structures.](https://pubmed.ncbi.nlm.nih.gov/35609995/) 6. [AlphaSync is an enhanced AlphaFold structure database synchronized with UniProt.](https://pubmed.ncbi.nlm.nih.gov/41219581/) 7. [M3Site: multiclass multimodal learning for protein active site identification and classification.](https://pubmed.ncbi.nlm.nih.gov/41222559/) 8. [Design and Validation of a Multi-Epitope mRNA Vaccine Construct Against Human Monkeypox Virus (hMPXV) by Annotating Protein of Intracellular Mature Virus (IMV) Form of hMPXV.](https://pubmed.ncbi.nlm.nih.gov/40564159/) 9. [GPCRchimeraDB: A Database of Chimeric G Protein-Coupled Receptors (GPCRs) to Assist Their Design.](https://pubmed.ncbi.nlm.nih.gov/40268234/)