# How to Identify an Unknown Protein Sequence Using UniProt's BLAST Tool By the end of this guide you will be able to take an unknown [amino acid sequence](/blog/guides/amino-acid-sequence), run it through the BLAST service at UniProt, read the hit list and the pairwise alignment correctly, decide whether the top hit actually identifies your protein, and record the accession number and organism with confidence. You need a web browser and your sequence in plain text. Everything else runs on the UniProt website, so there is nothing to install unless you later want the command-line alternatives described near the end. This workflow works for a peptide tag from mass spectrometry, a fragment you recovered from an expression clone, a translated open reading frame from a transcriptome assembly, or any amino acid string whose identity you need to pin down. ## What You Need Before You Start Gather these items first: 1. Your unknown sequence as amino acid single-letter code (A, R, N, D, C, Q, E, G, H, I, L, K, M, F, P, S, T, W, Y, V). If your sequence contains stop codons translated as asterisks, strip them before submitting. 2. Any context you have about the source: species, tissue, expected molecular weight, or the experiment that produced the fragment. Context is not required for the search, but it becomes valuable when you judge the hits. 3. A plain text editor (Notepad, TextEdit in plain text mode, VS Code, nano). Word processors insert hidden characters and smart quotes that break FASTA files. 4. Optionally, a spreadsheet or [lab notebook](/knowledge/molecular-biology/lab-notebook) to record accession numbers, E-values, and percent identity for each sequence you identify. A working definition of BLAST helps here. BLAST, short for Basic Local Alignment Search Tool, compares your query sequence against a database and reports local regions of alignment that score above a statistical threshold. For proteins, the program is called BLASTP, where P stands for protein. UniProt, the Universal Protein Resource, is a curated knowledgebase of [protein sequence](/blog/guides/protein-sequence) and functional information maintained through a collaboration that includes EMBL-EBI, and it exposes BLAST directly against its own protein sets. Knowing how to use UniProt well is therefore one of the highest-yield skills in routine sequence analysis, and the same pattern of search, inspect, and verify carries over to almost every UniProt database task you will attempt. ## Why UniProt BLAST Is a Good First Stop for Protein Identification Three properties make UniProt a strong first stop when you need to put a name on a sequence. First, the database is tiered by curation quality. UniProtKB, the core knowledgebase, contains two sections. Swiss-Prot holds entries reviewed by expert biologists with evidence-tracked annotation. TrEMBL holds unreviewed entries, mostly translated from coding sequences in the nucleotide databases and annotated by automated pipelines. When you BLAST against Swiss-Prot, every hit you see comes with human-reviewed functional annotation, which makes the jump from "similar sequence" to "plausibly this protein" much shorter. Second, UniProt BLAST gives you access to the UniRef clusters (UniRef100, UniRef90, and UniRef50), which collapse redundant sequences. Searching UniRef100 first can reveal the breadth of the protein family your query belongs to, then a follow-up search against Swiss-Prot pins down the best-annotated representative. Third, and most importantly for reliability, BLAST results are a starting point, not a verdict. A cautionary example comes from the literature on milk proteins. Researchers who had provisionally identified a protein as a human chorionic gonadotropin isoform re-examined the sequence by running BLAST searches, and follow-up pairwise alignments with NCBI Align, UniProt Align, and ExPASy Align showed the sequence was more similar to human bile salt-stimulated lipase than to hCG. The authors recommended the accession be relabeled as a partial bile salt-stimulated lipase [1]. That story captures the central lesson of this article: BLAST identifies candidates, and a careful analyst verifies them with independent checks before committing to a name. UniProt-based BLAST analysis is a standard component of published identification workflows across fields. Studies of parasite antigens have identified selenium-dependent glutathione peroxidase sequences using BLAST and UniProt databases together [2]. Epitope-mapping studies have performed [multiple sequence alignment](/blog/guides/multiple-sequence-alignment) analysis using BLAST and UniProt to position B-cell epitopes within a vaccine candidate protein [3]. Transcriptome projects routinely annotate assembled contigs by similarity searches against UniProt among other databases [4]. When you learn this workflow, you are learning the same move those papers make. ## Preparing Your Sequence BLAST is sensitive to junk in the input. Clean the sequence before you submit it. ### Format the sequence as FASTA FASTA format is a one-line header beginning with the greater-than symbol, followed by the sequence itself, usually wrapped across lines of 60 to 80 characters: ``` >unknown_protein_fragment | source: bovine milk fraction, ~87 kDa band DAHKSEVAHRFKDLGEENFKALVLIAFAQYLQQCPFEDHVKLVNEVTEFAK ``` The header text is for your records only. The search engine reads the amino acid letters. A sequence without a header line is also accepted by the UniProt BLAST form. ### Cleaning rules that prevent errors - Remove whitespace, digits, and punctuation from the sequence body. Copy-pasted mass spectrometry reports often contain line numbers, spaces, or tab characters. - Acceptable ambiguity codes are B, Z, X, and the stop symbol asterisk for some tools, but X is the only one you should tolerate in bulk. A run of X residues provides no alignment information, and long X stretches can push your query out of usable range. Replace uncertain residues with X one at a time rather than leaving blocks. - Remove leader sequences and vector tails only if you know they are artifacts. If the fragment came from a cloning experiment, trim the known vector-derived residues first, because a vector tail can dominate the top hits and bury the real protein. - Strip all numbering, such as residue position labels copied from a paper, and convert any one-letter lowercase letters to uppercase. Some BLAST interfaces treat lowercase as soft-masked (low-complexity) sequence, which changes the search behavior. ### Minimum length considerations Short queries are the most common cause of uninformative results. A 6-residue peptide will match thousands of unrelated proteins by chance. A 30-residue peptide with a good hit at high identity can be persuasive. As a practical planning rule, aim for at least 25 to 30 consecutive, confidently assigned residues before expecting a specific hit. If your only evidence is a shorter peptide tag, plan to use the hit list as a candidate pool for further testing rather than as an identification. ## UniProt BLAST Protein Sequence Identification Step by Step ![Flow from cleaning the sequence to saving the accession, with a chip showing the 50-residue example.](/article-images/illustrations/uniprot-blast-protein-sequence-identification-blast-workflow.webp) The following walkthrough uses the web form at the UniProt website. For the worked example, suppose your unknown fragment is this 50-residue peptide: ``` DAHKSEVAHRFKDLGEENFKALVLIAFAQYLQQCPFEDHVKLVNEVTEFAK ``` ### Step 1: Open the BLAST service Navigate to https://www.uniprot.org/blast. The form has a sequence input box on the left and a panel of database and parameter controls. If you are starting from the UniProt home page, BLAST is listed in the main navigation. This page is part of the broader UniProt interface, and the same account-free workflow applies across the site, which is one reason UniProt database tutorials start here. ### Step 2: Paste your sequence in FASTA format Paste the cleaned sequence into the input box. The form recognizes FASTA and plain sequence. After pasting, the interface reports the sequence length. For the example above, it should report a length of 50 amino acids. If the reported length is longer than the sequence you pasted, invisible characters came along with the copy. Paste into a plain text editor, delete anything that is not a valid amino acid letter, and paste again. ### Step 3: Choose the target database The database dropdown controls what your query is compared against. The main choices are: - **UniProtKB**, the complete knowledgebase including both Swiss-Prot and TrEMBL. Largest scope, most hits, includes automatically annotated entries. - **UniProtKB/Swiss-Prot**, the reviewed section only. Smaller, cleaner hit list, and the entries carry expert annotation. - **UniRef100, UniRef90, UniRef50**, clustered reference sets that merge identical and highly similar sequences at 100, 90, and 50 percent identity cutoffs respectively. These give a fast view of the sequence family and reduce redundancy in the hit list. For a first identification pass on an unknown protein, start with UniProtKB (or UniRef100 if your query may come from an organism with sparse coverage). Then run a second search against Swiss-Prot to find the best-annotated representative of the family. This two-pass approach mirrors published practice, where searches against UniProt databases are used both to find homologs and to anchor functional annotation [2]. ### Step 4: Choose the BLAST program UniProt offers BLASTP for plain protein-versus-protein comparison and PSI-BLAST, position-specific iterative BLAST, which builds a profile from the first round of hits and rescreens the database to pull in more distant relatives. For an unknown protein, start with plain BLASTP. Reserve PSI-BLAST for the second stage of a difficult identification, described later in this guide. ### Step 5: Set the parameters The defaults are sensible for most identification tasks. The parameters worth understanding and occasionally changing are: | Parameter | What it controls | Practical guidance | |, - |, - |, - | | Target database | Which sequence set is searched | UniProtKB for breadth, Swiss-Prot for curated annotation | | Program (BLASTP or PSI-BLAST) | Single-pass search or iterative profile search | BLASTP first, PSI-BLAST only after judging round one | | E-value threshold | Statistical cutoff for reporting hits | Default around 0.1 on the web form, tighten to 1e-5 or lower for confident calls | | Matrix (BLOSUM62 default) | Scoring of amino acid substitutions | BLOSUM62 for general searches, BLOSUM45 for distant relationships, BLOSUM80 for short similar queries | | Number of alignments shown | Size of the returned hit list | Increase to 100 or more if the top region of the list is crowded | | Low-complexity filter | Masks biased regions such as coiled coils and low-complexity tails | Leave on for long queries, consider off for short queries where the signal itself is low complexity | | Organism restriction | Limits hits to selected taxa | Use when source species is known to cut noise, avoid when identifying from an unknown source | The E-value deserves a definition. It is the expected number of database matches with a score at least as good as the one observed that would occur by chance alone in a database of the size searched. An E-value of 1e-50 means chance would produce a match this good roughly once per 10 to the 50th power searches, which is effectively a certain homolog. An E-value of 0.5 means matches this good appear by chance about half the time, which is noise. Between those extremes lies the gray zone where judgment matters, addressed below. ### Step 6: Run the search and read the hit table Click Run. The search against UniProtKB typically completes in well under a minute for a query of ordinary length. The results page opens with a graphical summary of the aligned regions and then a table of hits. Each row reports the entry accession, entry name, organism, alignment length, percent identity, and E-value. For the 50-residue albumin-fragment example, a representative top of the hit list against UniProtKB looks like this (values shown are typical of a near-exact match, your exact numbers depend on the current database release): ``` Accession Entry name Organism Identity Length E-value P02768 ALBU_HUMAN Homo sapiens 100.0% 50 2e-30 P43652 ALBU_FELCA Felis catus 84.0% 50 8e-22 P02769 ALBU_BOVIN Bos taurus 76.0% 50 3e-19 ... ``` Reading that table, the call is immediate. A 100 percent identical 50-residue match to human serum albumin, with an E-value many orders of magnitude below 1e-5, identifies the fragment as albumin. The next rows are orthologous albumins from related mammals at lower identity, which is exactly the pattern a genuine albumin fragment should produce, and that pattern is itself a piece of evidence. A real identification usually shows a gradient of related hits beneath the top hit, not a lone match followed by nothing. ### Step 7: Open the alignment and judge the match Click the top hit to expand the pairwise alignment. Three numbers govern how much weight the alignment deserves: 1. **Percent identity**, the fraction of aligned positions with identical residues. Above roughly 30 percent identity over a substantial length, homology is usually secure, and the higher the value, the more confidently you can transfer functional annotation from the hit to your query. This is textbook territory, and it aligns with what function-prediction studies have shown quantitatively: enzyme function prediction by BLAST becomes unreliable when no homolog with identity above 30 percent is available, which is why dedicated domain-based methods were developed for that regime [5]. 2. **Coverage**, the fraction of your query that participates in the alignment. A 100 percent identity match that covers only 12 residues of a 400-residue query identifies the fragment, not the whole protein. Always check that alignment length is close to query length when your goal is full-protein identification. 3. **E-value and bit score**, jointly summarizing statistical strength. Within a hit list, prefer the hit with the best combination of all three measures rather than percent identity alone. A 95 percent identity match over 40 residues can carry a worse E-value than an 80 percent identity match over 300 residues, and the longer alignment is usually the more informative one. Also inspect where mismatches fall. A handful of conservative substitutions spread along the alignment supports an ortholog call. A cluster of differences at one end can indicate a [signal peptide](/knowledge/molecular-biology/signal-peptide) or isoform boundary, or a sequence error at one end of your fragment. ### Step 8: Confirm the call with a second method Before you write the identification into a notebook or manuscript, verify it independently. The hCG versus bile salt-stimulated lipase case was resolved only after the original BLAST-based call was tested with several pairwise alignment methods, and only the combined evidence settled the matter [1]. Practical verification steps: - **Reciprocal BLAST.** Copy the best hit's full sequence from UniProt and run it back through BLAST restricted to organisms and databases relevant to your source. A genuine identification reciprocates: the original hit should match your query region as its best match in the reciprocal direction. - **Pairwise alignment.** Use the Align tool on UniProt entry pages or a service such as ExPASy Align to align your query directly against the candidate, without the database search in between. Agreement between database BLAST and direct pairwise alignment removes the possibility that the hit table ranked a misleading partial match first [1]. - **Multiple sequence alignment.** If several hits tie, align your query with the top candidates plus known family members. This is the approach used in published epitope-mapping work, where BLAST and UniProt-based multiple sequence alignment placed the mapped epitope in context across homologs [3]. A correct identification sits cleanly inside its family, with gaps that make structural sense. - **Domain check.** Open the candidate UniProt entry and compare its annotated domains and features against what your fragment covers. If the candidate is an enzyme with a catalytic site at residues 120 to 125 and your query covers that region with perfect identity, your confidence rises. If your query covers only a low-information tail, confidence drops regardless of the percent identity. ### Step 9: Retrieve the entry and record the identifiers Once convinced, record the UniProtKB accession (for example P02768), the entry name, the organism, the reviewed status, and the database release or access date. UniProt entries change as annotation is updated, so the accession plus the date is what makes your identification reproducible. If you need the full sequence of the hit for the reciprocal check or for primer and peptide design, download it directly with the UniProt REST API: ``` curl "https://rest.uniprot.org/uniprotkb/P02768.fasta" -o albumin_human.fasta ``` And if you want to pull a set of candidate entries in a tabular format for a notebook: ``` curl "https://rest.uniprot.org/uniprotkb/search?query=protein_name:serum+albumin+AND+organism_id:9606&format=tsv&fields=accession,id,protein_name,organism_name,length" > albumin_candidates.tsv ``` That search endpoint accepts the same query syntax as the UniProt website search box, which is a fact worth internalizing, because it means anything you can find by browsing you can also fetch programmatically. ## Key Parameters at a Glance The table above lists the adjustable parameters. One more decision deserves its own treatment because it changes the character of the search: the choice among UniRef cluster sizes. - **UniRef100** merges fully identical sequences and fragments. Use it when you want the closest possible matches and maximal organism coverage. - **UniRef90** merges sequences that are at least 90 percent identical. It shortens the hit list while preserving near-species-level resolution. - **UniRef50** merges at 50 percent identity and is designed for family-level questions and for speeding up downstream searches. It answers "what family does this belong to" faster than it answers "what exact protein is this." A sensible identification routine is UniRef100 or UniProtKB for the first pass, then Swiss-Prot for the annotated representative, then reciprocal and pairwise verification. Projects that build databases from many sequences follow the same funnel logic, clustering with bidirectional best BLAST hits and then consolidating families, as done for the bacterial small-[protein database](/blog/guides/protein-database) sORFdb [6]. ## Interpreting Scores: E-Value, Identity, and Coverage in Practice ![Three yes-or-no questions on identity, coverage and hit-list shape leading to a final call to name or keep as candidate.](/article-images/illustrations/uniprot-blast-protein-sequence-identification-hit-triage.webp) New analysts tend to over-weight percent identity and under-weight coverage and alignment length. Consider three hypothetical outcomes for a 300-residue query: 1. Hit A, 98 percent identity over 280 residues, E-value 1e-150. Near-certain identification of the same protein or a very recent ortholog. 2. Hit B, 100 percent identity over 25 residues, E-value 3e-8. The E-value is formally significant, but a 25-residue perfect match recurs by chance far more often than a 280-residue match at 98 percent. Treat it as a candidate, not an identification. 3. Hit C, 34 percent identity over 300 residues, E-value 1e-12. Homology is probably real, but naming the protein precisely from this hit alone is risky. The literature on enzyme annotation quantifies this risk: when the best available homolog sits near or below 30 percent identity, BLAST-based functional assignment loses accuracy, and methods that combine domain signatures outperform BLASTP in that regime [5]. In other words, Hit C tells you the family, not the function. A second pattern to recognize is the lone strong hit with no supporting relatives. If the top hit claims a specific protein at high identity but the rest of the list is filled with unrelated low-scoring alignments, suspect a database anomaly, a chimeric database entry, or a contaminant in your sample that happens to match. Contrast that with the expected gradient, where dozens of orthologs and paralogs trail the top hit at decreasing identity. Finally, remember that the database itself can be wrong. The bile salt-stimulated lipase episode arose because a database entry carried a misleading functional label, and BLAST plus alignment evidence was needed to correct it [1]. UniProt reviewed entries are the most reliable annotations in public protein science, but every database contains some misannotated records, and your job as the analyst is to weigh evidence, not to defer to a label. ## Using PSI-BLAST for Distant Homologs When plain BLASTP returns only weak hits or nothing above the noise floor, PSI-BLAST is the escalation tool. It runs an initial BLASTP, builds a position-specific scoring matrix from the aligned hits, and then re-searches the database using that profile, typically for one to three iterations. Distant homologs that a single-sequence search misses often appear in iteration two. A published example shows exactly when to reach for it. In the hunt for a selenium-dependent glutathione peroxidase in the filarial worm [Setaria digitata](/knowledge/parasites/livestock-parasites/setaria-digitata-cattle-horses-cerebrospinal-nematodiasis), the limited number of nematode entries in the databases made direct searching unreliable, so the researchers used a validated SeGPx sequence from a related organism, the great pond snail Lymnaea stagnalis, as a proxy query in PSI-BLAST to identify homologs, and then confirmed the gene's location in the target genome and the enzyme's presence and activity in worm extract [2]. Two lessons transfer directly to any identification task: 1. When your source organism is poorly represented in the databases, a well-annotated homolog from a related species is a legitimate query. The identification then proceeds through family membership rather than through a direct species match. 2. Iterative profile searches widen the net, and a computational identification gains weight when an independent line of evidence, such as an enzymatic assay in that study, supports it [2]. PSI-BLAST has one famous failure mode. Every iteration adds hits to the profile, and if a marginal or false hit enters the profile, subsequent iterations chase it and pull in more false positives, a process called profile drift. Guard against it by checking every hit you leave in the profile after each iteration, and by stopping once the list stabilizes. For identification of an unknown protein, one or two iterations are usually sufficient. ## Going Past the Web Form: Programmatic and Command-Line Routes The UniProt web form is ideal for tens of sequences. For larger workloads, two routes keep the same database in play. **UniProt REST API for retrieval.** The BLAST service itself is driven by the website, but every entry and search result is retrievable over the REST API, so a common pattern is to run BLAST once, harvest the accession list, and then batch-download entries and annotations: ``` curl "https://rest.uniprot.org/uniprotkb/search?query=accession:(P02768 OR P02769 OR P43652)&format=tsv&fields=accession,id,protein_name,organism_name,length,reviewed" > candidates.tsv ``` **Local BLAST against UniProt data.** Download a Swiss-Prot FASTA file from the UniProt download page, format it with the NCBI BLAST+ toolkit, and search locally. The standard invocation is: ``` blastp -query unknown.fasta -db swissprot -evalue 1e-5 \ -outfmt "6 sseqid stitle pident length evalue bitscore" \ -num_threads 8 > hits.tsv ``` This is the same calculation the web form performs, with full control over parameters and the ability to script thousands of queries. Larger database-construction projects operate at exactly this level, using BLAST searches against UniProt and related sources as the annotation backbone for their systems, as the Spodoptera EST database did with its annotation against NCBI and UniProt and integrated BLAST search access [7], and as the oak transcriptome study did when it assigned functional descriptions to assembled transcripts by searching UniProt among seven public databases [4]. When your identification task grows past a handful of sequences, that is the infrastructure to imitate. One more pattern from the literature is worth knowing. Interaction-prediction servers use BLAST to map a query sequence onto structural templates and then return candidate partner proteins as UniProt identifiers [8]. If your unknown protein turns out to be a known family member, those same UniProt identifiers become entry points into interaction and structure resources, which is a useful next step once the identification is settled. ## Common Mistakes and Limitations These are the errors that most often produce a wrong or meaningless protein identification. 1. **Treating the top hit as the answer.** The top hit is a candidate. The bile salt-stimulated lipase case began with a database-based call that only failed open when someone looked closer and ran confirmatory alignments [1]. Always check identity, coverage, and the shape of the hit list before naming. 2. **Searching TrEMBL and trusting the functional label.** TrEMBL entries are annotated by automated similarity transfer, so a wrong label can propagate from entry to entry. When an important call rests on an unreviewed entry, find the reviewed Swiss-Prot member of the family and read its evidence. 3. **Confusing a fragment match with a protein identification.** A 20-residue perfect match inside a 500-residue query tells you your protein contains that region. It does not tell you the rest of the protein. Check that alignment length covers the meaningful part of your query. 4. **Naming a function at low identity.** Below roughly 30 percent identity to any homolog, BLAST alone cannot carry a functional assignment, a limitation demonstrated quantitatively in benchmarked enzyme-prediction work [5]. Report the family, not the enzyme name, unless another method such as domain analysis supports the call. 5. **Ignoring the hit-list shape.** Missing relatives under the top hit, as described above, signal that something is off. Real homologs come with families. 6. **Letting PSI-BLAST drift.** Accepting every automatic hit into the profile during iteration imports false positives that compound. Curate each iteration manually. 7. **Expecting hits for very small proteins.** Small proteins of fewer than 100, and especially fewer than 50, amino acids are systematically underrepresented and under-annotated in databases, and many lack known homologs entirely, which was a central finding motivating the sORFdb resource [6]. A short query that returns nothing may be a genuine novel small protein rather than a bad sequence. 8. **Submitting dirty input.** Numbers, spaces, smart quotes, asterisks, and lowercase letters copied from documents are the most common causes of immediate errors or silent length mismatches. Clean to pure uppercase single-letter code. 9. **Forgetting that the database changes.** UniProt is updated continuously. Accessions are stable, but names, annotations, and hit rankings shift between releases. Record the access date with every identification. 10. **Over-reading a match to a hypothetical or predicted protein.** A hit labeled "hypothetical protein" or "predicted" transfers no functional information. BLAST has established that statistically significant similarity to UniProt entries can confirm coding potential and family membership for novel predicted sequences [9], but family membership is where the information stops until reviewed annotation or experimental evidence exists. On limitations more broadly: an in silico identification is a hypothesis about molecular identity, and any conclusion that feeds into downstream biological work deserves confirmation by an independent method, whether that is targeted mass spectrometry, enzymatic activity as in the SeGPx study [2], or immunological detection. ## Frequently Asked Questions ### What E-value should I trust when identifying a protein? Trust depends on alignment length as much as the E-value itself. As a working rule, an E-value at or below 1e-5 for an alignment covering most of your query is a credible homolog, values many orders of magnitude lower with high identity support a specific identification, and anything between 1e-3 and 0.1 should be treated as a candidate requiring verification. ### Should I search Swiss-Prot or the full UniProtKB? Search UniProtKB first for breadth, especially if your source organism may be poorly covered, then repeat the search against Swiss-Prot to find the best-annotated representative of the family. Swiss-Prot entries carry expert-reviewed annotation with evidence tags, which is what you want to anchor your final call. ### Why did UniProt BLAST return no hits? The most frequent causes are a very short query, a sequence with many ambiguous or nonstandard characters, a genuinely novel or underrepresented protein family, or a query from an organism with sparse database coverage. Clean the input, extend the sequence if possible, and if the search still fails, try a homolog from a related organism as a proxy query as done in published work on poorly covered species [2]. ### What is the difference between UniRef100, UniRef90, and UniRef50? They are clustered versions of the protein space merged at 100, 90, and 50 percent identity. UniRef100 preserves maximum sequence detail, UniRef90 trims near-duplicates, and UniRef50 collapses each family to representative sequences for fast family-level questions. For exact identification, use UniProtKB or UniRef100, and use UniRef50 when you only need to know the family. ### Can I paste a DNA sequence into UniProt BLAST? The UniProt BLAST service compares protein sequences against protein databases, so paste amino acids. Translate your DNA in all six reading frames first, either with a translation tool or with software such as blastx locally, then submit the resulting protein sequence. Submitting raw nucleotide letters to a protein search produces meaningless alignments. ### How much sequence identity do I need to name a protein? Homology becomes generally secure above roughly 30 percent identity over a substantial aligned length, but precise naming and reliable functional transfer require much higher identity, and below that 30 percent line BLAST-based function calls become unreliable, a regime where domain-based methods perform better [5]. For a confident species-level identification, expect high identity over most of the query length. ### What does "unreviewed" mean in UniProtKB results? It marks a TrEMBL entry, annotated automatically by similarity and rule-based pipelines rather than by a human curator. The sequence itself is usually fine, but the functional annotation is a computational inference. Treat unreviewed labels as leads and confirm important calls against the reviewed Swiss-Prot member of the family. ### Can UniProt BLAST identify a protein from a short mass spectrometry peptide tag? It can return the candidate pool, and a peptide tag of 25 to 30 residues with a unique high-identity hit can be persuasive. Shorter tags match many unrelated proteins by chance, so use the hit list to generate candidates and confirm with additional evidence such as a second peptide, mass spectrometric fragment matching, or a targeted experiment. ## Sources 1. [Identification of an hCG Isoform Which Is Similar in Sequence to a Bile Salt-Stimulated Lipase.](https://pubmed.ncbi.nlm.nih.gov/42907636/) 2. [Sequence Localization of SeGPx in S. digitata Genome Contigs and Determination of its Presence in the Whole Worm Extract.](https://pubmed.ncbi.nlm.nih.gov/40694172/) 3. [Identification of B-cell dominant epitopes in the recombinant protein P29 from Echinococcus granulosus.](https://pubmed.ncbi.nlm.nih.gov/35478448/) 4. [RNA-seq analysis of Quercus pubescens Leaves: de novo transcriptome assembly, annotation and functional markers development.](https://pubmed.ncbi.nlm.nih.gov/25393112/) 5. [DomSign: a top-down annotation pipeline to enlarge enzyme space in the protein universe.](https://pubmed.ncbi.nlm.nih.gov/25888481/) 6. [sORFdb - a database for sORFs, small proteins, and small protein families in bacteria.](https://pubmed.ncbi.nlm.nih.gov/39910485/) 7. [SPODOBASE: an EST database for the lepidopteran crop pest Spodoptera.](https://pubmed.ncbi.nlm.nih.gov/16796757/) 8. [HOMCOS: a server to predict interacting protein pairs and interacting sites by homology modeling of complex structures.](https://pubmed.ncbi.nlm.nih.gov/18442990/) 9. [Classification analysis of triplet periodicity in protein-coding regions of genes.](https://pubmed.ncbi.nlm.nih.gov/18593596/)