# UniProt Cross-References for Non-Model Organisms The one-sentence difference is this: for model organisms, UniProt cross-references already exist as curated links you simply read off the entry, while for non-model organisms you usually have to build those links yourself, either by mapping identifiers between databases or by mapping sequences at 100 percent identity. The rule of thumb for choosing between them is straightforward. If you hold a stable identifier that UniProt recognizes, such as a UniProtKB accession or a RefSeq protein ID, use the UniProt ID mapping service or the cross-reference section of the entry itself. If all you have is a raw sequence, or if the identifier refuses to map, route the problem through UniParc, the comprehensive sequence archive, and demand an exact sequence match before you accept any cross-reference [1]. This article is a practical UniProt database tutorial focused on the species most bioinformaticians actually work on, which are often organisms with no curated model-organism database behind them. It walks through every option for building and using cross-references, compares them on the criteria that decide real choices, works through concrete scenarios, and ends with the mistakes that cost people the most time. ## What a UniProt Cross-Reference Actually Is UniProt is the central protein knowledgebase, and UniProtKB is its core database of protein entries. UniProtKB has two divisions. Swiss-Prot contains reviewed entries that human curators have annotated and checked. TrEMBL contains unreviewed entries annotated by automated pipelines. For most non-model organisms, nearly everything you will touch lives in TrEMBL. Every UniProtKB entry carries an accession, a stable identifier of the general form P12345 for reviewed entries or a letter-and-number pattern such as A0A023GPI8 for many TrEMBL entries. Alongside the sequence and annotation, each entry carries a block of cross-references. In the flat-file format these appear as DR lines, and on the website they appear under the heading "Cross-references." A cross-reference is a pointer from the UniProt entry to the same protein, or a closely related record, in another database. Typical targets include RefSeq at NCBI, GeneID at NCBI, Ensembl, the [Protein Data Bank](/blog/guides/protein-data-bank), KEGG, OrthoDB, the Gene Ontology Annotation resources, and domain databases such as Pfam and InterPro. A concrete example makes the shape clear. A well-annotated human entry might carry DR lines pointing to dozens of external resources, giving you a one-click bridge from a UniProt accession to RefSeq, PDB structures, pathway databases, and literature. A TrEMBL entry for a protein from a recently sequenced parasite or an orphan fungal genome might carry only a handful of cross-references, often just Ensembl Genomes, GeneID, and a few domain predictions. The cross-reference infrastructure is identical in both cases. The density and reliability of the links is not. ## Why Non-Model Organisms Break Simple Cross-Referencing Identifier mapping is a solved problem for human, mouse, rat, zebrafish, fruit fly, worm, yeast, and Arabidopsis, because years of curation have produced dense, maintained links between UniProt, NCBI, Ensembl, and the model-organism databases. Outside that circle, three problems dominate. First, identifiers are unstable. As Côté and colleagues framed the problem in their description of the PICR service, each major [protein database](/blog/guides/protein-database) uses its own conventions when assigning protein identifiers, and resolving the various, potentially unstable, identifiers that refer to identical proteins is a major challenge when you try to unify datasets annotated against different sources [1]. A genome reannotation can retire thousands of locus tags. RefSeq proteins get merged or suppressed. TrEMBL accessions occasionally change when the underlying nucleotide record changes. Second, coverage is asymmetric. Your species may have RefSeq proteins but no Ensembl Genomes entry, or the reverse. Gene Ontology annotations may exist only by electronic inference. A mapping service that works beautifully for Escherichia coli may return nothing at all for a newly sequenced annelid. Third, partial solutions abound. As the PICR authors noted, the mapping tools available before their work were limited to specific species or techniques and to a small number of databases, which is exactly the situation a non-model organism researcher faces when a pipeline built for mouse quietly fails on a coral [1]. These three problems explain why sequence identity, rather than identifier linkage, becomes the anchor of trust in non-model work. Two records that share an identical [amino acid sequence](/blog/guides/amino-acid-sequence) are the same protein at the sequence level, regardless of what any database calls them. ## The Core Comparison: UniProt Cross-Reference Options Side by Side The table below is the backbone of this article. Each row is a distinct option for obtaining or using cross-references, and each is explained in its own section afterward. | Option | What it maps | What you need as input | Identity guarantee | Best for | Main limitation | |--|--|--|--|--|--| | Cross-reference section on a UniProtKB entry | One entry to its linked records | A UniProtKB accession or a search that lands you on the entry | Curator-maintained or automatic, varies | Quick lookups, model-organism-style browsing | Sparse for non-model species, updates lag | | UniProt ID mapping service | Identifier sets between about a hundred database namespaces | Recognized identifiers in a source database | Namespace-level mapping, not guaranteed sequence identity | Bulk conversion of gene or protein ID lists | Silent drops when an ID has no counterpart | | UniParc lookup and PICR | Identical sequences across more than 70 source databases [1] | Sequences in FASTA format, or protein identifiers, or UniParc entries | 100 percent sequence identity by design [1] | Identifier archaeology, unifying datasets across sources, unstable IDs | Identity says nothing about annotation quality | | UniProt BLAST | Your sequence to the closest sequences in UniProtKB | A protein or [nucleotide sequence](/knowledge/molecular-biology/nucleotide-sequence) in FASTA format | Similarity, not identity, unless hits are exact | Non-model species with no IDs at all, transcript-to-protein assignment | Best hit is not always ortholog, needs manual sanity checks | | UniProt REST API | Any of the above, programmatically | A script and HTTP access | Inherits the guarantee of the underlying service | Pipelines, hundreds to millions of identifiers | Requires polling for asynchronous jobs, rate limits | | Batch downloads and custom mapping | Whole proteomes, whole database slices | Internet bandwidth and a scripting language | Whatever you enforce in your script | Very large or repeated analyses, full reproducibility | You own the bookkeeping and the updates | ## Option 1: Reading the Cross-Reference Section on an Entry The simplest way to use UniProt is to arrive at an entry and read its cross-references. Search UniProtKB for your protein name plus your organism name, open the entry, and look for the "Cross-references" section on the left-hand menu of the entry page. In the text format of the flat file, the same information appears as DR lines near the top. For a non-model organism, work the section in a fixed order. Look first for cross-references to sequence databases, such as RefSeq, GeneID, and Ensembl Genomes, because those give you alternative identifiers for the same protein. Look second for cross-references to pathway and function databases, such as KEGG, Reactome, Gene Ontology, and InterPro, because those give you the functional context you usually wanted in the first place. Look last for cross-references to literature and structure databases. The pros of this route are zero setup and curator context. The cons are that it works one entry at a time, that non-model entries often have thin cross-reference blocks, and that links can lag behind the source databases by months. If you need more than a few dozen proteins, skip straight to the ID mapping service. ## Option 2: The UniProt ID Mapping Service The ID mapping service, reachable from the "ID Mapping" link in the tools area of uniprot.org, converts a list of identifiers from one database namespace into another. It supports a large set of source and target databases, including UniProtKB accessions, RefSeq, GeneID, Ensembl and Ensembl Genomes, PDB, KEGG, UniParc, and many more. This is the tool behind most answers to the question of how to use UniProt for bulk identifier conversion. The workflow is the same on the website and through the API. You declare a source database, a target database, and a list of identifiers. The service resolves each identifier to UniProtKB entries, then reads the cross-references on those entries to produce the target identifiers. A concrete worked example on the website: 1. Open the ID mapping page. 2. Set "From database" to RefSeq and "To database" to UniProtKB. 3. Paste identifiers such as WP_003197841, one per line. 4. Submit, wait for the job to finish, and download the mapped pairs as tab-separated text. The same job through the REST API, which you would embed in a pipeline: ``` curl -X POST "https://rest.uniprot.org/idmapping/run" \ -d "from=RefSeq" \ -d "to=UniProtKB" \ -d "ids=WP_003197841" ``` The response contains a job identifier. Poll the status endpoint until the job state reads FINISHED, then stream the results: ``` curl "https://rest.uniprot.org/idmapping/stream/REPLACE_WITH_JOB_ID" ``` The pros are bulk throughput, official support, and reproducibility through the API. The cons matter for non-model organisms in three specific ways. Mapping can fail silently, meaning an identifier with no counterpart simply returns no row, so always compare input and output counts. Mapping can expand, meaning one input identifier returns several output identifiers, which happens constantly with TrEMBL entries that share sequence identity with reviewed entries. And the service maps by cross-reference, not by sequence comparison, so a mapped pair reflects what the databases claim, not what the sequences prove. ## Option 3: UniParc and PICR, the Sequence-Identity Route UniParc, the UniProt Archive, is the warehouse that makes sequence-based cross-referencing possible. It stores every distinct [protein sequence](/blog/guides/protein-sequence) ever loaded into UniProt from its source databases, without annotation and without organism-specific duplicates, and gives each unique sequence a stable UniParc identifier beginning with UPI. When the same sequence appears in RefSeq, Ensembl, and UniProtKB, all of those records point at the same UniParc entry, and the UniParc entry lists every source database that contains the sequence, including entries that have since become inactive or obsolete. PICR, the Protein Identifier Cross-Referencing service, is the web application and programmatic service built on top of that warehouse. According to its authors, PICR provides interactive and programmatic access, through SOAP and REST interfaces, to a mapping algorithm that uses UniParc as a data warehouse to offer cross-references based on 100 percent sequence identity to proteins from over 70 distinct source databases loaded into UniParc [1]. Three design choices make PICR unusually useful for non-model organisms. Mappings can be limited by source database, so you can request only RefSeq or only Ensembl targets. Mappings can be limited by taxonomic identifier, so a coral protein will not return a human hit. Mappings can be limited by activity status in the source database, so you can suppress retired identifiers and keep only records that are currently live [1]. Users can copy and paste or upload files containing protein identifiers or sequences in FASTA format, and results can be viewed as simple or detailed HTML tables or downloaded in standard formats [1]. For a non-model organism, this route answers questions the other options cannot. When a locus tag from a 2014 genome assembly no longer resolves anywhere, UniParc can often still locate the sequence and enumerate where that sequence lives today. When a collaborator sends you identifiers from a database you have never heard of, PICR can translate them through the shared sequence rather than through any bilateral agreement between databases. A practical UniParc workflow through the web interface: 1. Retrieve your protein sequence in FASTA format. 2. Go to the UniParc search on uniprot.org and search with the sequence, or compute the sequence checksum locally and search on the checksum, which is a 32-character hexadecimal fingerprint of the exact sequence. 3. Open the matching UniParc entry and read the source-database table, which shows every database containing that sequence and whether each source entry is still active. The pros are the strongest correctness guarantee available, the widest database coverage of any single service, and taxonomic and activity filters that keep result lists sane. The cons are that exact identity is a strict bar, so sequences with a single residue difference will not map, and that identity carries no functional information, a point developed in the limitations section. ## Option 4: UniProt BLAST When You Have No Identifier At All Many non-model projects start with sequences and nothing else. A [de novo transcriptome assembly](/knowledge/bioinformatics/transcriptome-assembly-without-a-reference-genome) produces thousands of transcripts with arbitrary names like c12345_g2_i1. A new genome produces predicted proteins named by an annotation pipeline rather than by any public database. In these situations no identifier mapping service can help, because your identifiers exist only inside your project. The bridge to UniProt is sequence search. The UniProt BLAST service, linked from the "BLAST" item in the tools area of uniprot.org, takes a FASTA sequence and searches it against UniProtKB, against Swiss-Prot alone, against a proteome you specify, or against UniRef clusters. For non-model organisms, two settings matter more than the rest. First, restrict or at least label the search by taxonomy. The web form accepts an organism filter, so you can search only within your organism's proteome when one exists, or search all of UniProtKB and then filter results by taxon. Searching an orphan fungal protein without a taxon restriction invites you to interpret hits from distant species as if they were orthologs. Second, decide what the hit gives you. If your transcript matches a UniProtKB entry at 100 percent identity over its full length, you have effectively found the UniProt record for your protein, and you can now use every cross-reference on that entry. If the best hit is 34 percent identity over 60 percent of the length, you have found a distant relative, and the honest conclusion is functional inference by homology, not identifier mapping. Keep the two conclusions in separate columns of your results table. The pros are universal applicability, since every project has sequences even when it has no recognized IDs, and direct routing into the annotation system of UniProtKB. The cons are that BLAST returns similarity rather than identity, that best hits between non-model proteins and model-organism proteins can be paralogs rather than orthologs, and that BLAST is slower and more subjective than exact-match mapping, so treat it as hypothesis generation. ## Option 5: Programmatic Access for Pipelines Everything described so far has a programmatic interface, which matters because non-model organism analyses rarely involve a handful of proteins. The UniProt REST API at rest.uniprot.org supports entry retrieval with rich field selection, the asynchronous ID mapping job pattern shown in Option 2, UniParc queries, proteome downloads, and streaming results in TSV, JSON, or FASTA. The EBI Proteins API provides a complementary set of endpoints for cross-references and annotations. PICR adds its own SOAP and REST programmatic interfaces on top of UniParc [1], which means a script can submit a FASTA file and receive cross-references against dozens of source databases in one call. A minimal reproducible pattern for batch work: 1. Build your identifier list as a plain text file, one identifier per line, and archive it with your analysis. 2. Submit it to the ID mapping service through the API, capturing the job identifier in your log. 3. Poll the status endpoint, then stream the mapped results to a file. 4. Record the date, the source and target databases, and the API response in your project notes, because mapping results change as the underlying databases change. For sequence-based mapping at scale, the same pattern applies to PICR. Submit sequences in FASTA format, request a specific target source database and taxon, and request only active entries if you want current identifiers [1]. The pros are throughput, reproducibility, and the ability to enforce filters consistently. The cons are that asynchronous APIs require careful job handling, that undocumented rate limits can throttle very large submissions, and that scripting shifts the burden of error handling onto you. A mapping service returns an empty result for a failed identifier, and only your code will notice. ## Option 6: Batch Downloads and Custom Mapping Tables The heaviest-duty option is to stop querying services at all. UniProt provides complete proteome sets per organism, full UniProtKB releases, and the complete UniParc sequence archive as downloadable files. For a large or repeated analysis, download the proteome FASTA for your organism, download the corresponding table of identifiers and cross-references in TSV format, and build your own mapping table keyed on sequence checksums. This is essentially what PICR does internally, generalized to your own files. Because UniParc indexes sequences by exact content, you can join any two protein datasets, from any source databases, on the shared sequence, which is the same 100 percent identity logic that underpins PICR's mapping algorithm [1]. The difference is control. A custom table can carry your own confidence columns, your own release numbers, and your own orthology calls. The pros are total reproducibility, freedom from network calls during analysis, and the ability to handle millions of sequences. The cons are that you inherit the maintenance burden, that download files are large, and that a hand-rolled mapping script is the single most common source of silent mapping errors in published pipelines. ## Choosing Between the Options: The Criteria That Decide Real Choices ![Decision path asking whether you hold a recognized identifier, then whether it maps, ending at ID mapping, UniParc or BLAST.](/article-images/illustrations/uniprot-cross-references-non-model-organisms-route-choice.webp) When people ask about the best UniProt cross-references for non-model organisms, the honest answer is that the best option is a function of what you hold in your hands. Four criteria decide it. **Correctness guarantee.** Exact sequence identity is the strongest guarantee. PICR through UniParc applies a 100 percent sequence identity requirement across more than 70 source databases [1]. The ID mapping service applies whatever the source databases assert about their links, which is usually correct but unverified at the sequence level. BLAST applies similarity, the weakest guarantee, and requires human judgment. **Coverage for your organism.** Curated cross-references are dense for model organisms and thin elsewhere. Identifier mapping works only when the source and target databases both know about your species. UniParc coverage extends to anything loaded into it from its source databases [1], which is far broader. BLAST covers everything with a sequence. **Identifier stability.** If your input identifiers are versioned and current, any option works. If your identifiers are old, retired, or from an unversioned local assembly, sequence-based routes are the only dependable ones, and the activity-status filter in PICR exists precisely to separate live identifiers from dead ones [1]. **Throughput and reproducibility.** Reading entries works for tens of proteins. The ID mapping API works for thousands to millions. Batch downloads work for everything, forever, but only if you record release dates. Translated into a decision sequence, for any non-model organism job: try the ID mapping service first if your identifiers are recognized. Fall back to UniParc or PICR with a taxon filter when identifiers fail or when you need proven identity. Use BLAST to enter the UniProt system when you have sequences but no IDs at all. Automate through the REST API once the manual path is proven. ## Worked Scenarios ### Scenario 1: Annotating a de novo transcriptome from an orphan species You assembled a transcriptome from a non-model mollusk, translated the open reading frames, and now hold 20,000 protein sequences with names like TRINITY_DN1000_c0_g1_i1. No public database knows these names, so identifier mapping is impossible. Route A is to BLAST every sequence against Swiss-Prot, restrict attention to hits with high coverage and strong identity, and treat the matched UniProtKB accessions as your bridge to Gene Ontology terms, domains, and pathways, all of which are present as cross-references or annotation on those entries. Route B, for transcripts that should exactly match a protein in a source database, is PICR with the sequences in FASTA format and a taxonomic restriction to your organism's NCBI taxon identifier, which returns cross-references at 100 percent identity where they exist [1]. Use Route B for exact matches and Route A for everything else, and keep the two result classes labeled differently downstream. ### Scenario 2: Converting Ensembl Genomes identifiers to RefSeq for a parasite Your collaborator works with a parasitic nematode and sends a gene list in Ensembl Genomes protein identifiers. You need RefSeq accessions to feed a downstream tool. A single two-step job through the ID mapping service handles this. First map Ensembl Proteins to UniProtKB. Then map the resulting UniProtKB accessions to RefSeq. Expect expansion at both steps, because one Ensembl protein can correspond to several UniProtKB entries when isoforms or duplicate sequences exist, and one UniProtKB entry can link to several RefSeq records. For a short list, verify a sample of ten mappings by comparing the actual sequences, either visually or with an alignment tool. For a long list, apply the strongest available check instead and confirm the mappings through UniParc at 100 percent sequence identity, since UniParc sits underneath both databases and lists where each identical sequence appears [1]. ### Scenario 3: Resurrecting identifiers from a superseded genome assembly ![Flowchart from a dead locus tag to a UniParc search, then to active or inactive source records and current identifiers.](/article-images/illustrations/uniprot-cross-references-non-model-organisms-uniparc-rescue.webp) A 2016 paper on an orphan crustacean reports protein identifiers that no longer resolve anywhere. The assembly was reannotated, the locus tags were retired, and the identifier mapping service returns nothing. This is the canonical identifier archaeology problem, and UniParc exists for it. If you have the protein sequences, either from the old paper's supplementary files or by retranslating the old assembly, submit them to PICR, which maps through UniParc at 100 percent identity and can restrict results to currently active entries so you receive only identifiers that resolve today [1]. If you have only the old identifiers, search UniParc directly for them, because UniParc retains cross-references to source entries even after those entries become inactive in their source database [1]. The inactive records tell you what the old identifiers referred to, and the same UniParc entry tells you which current records carry the identical sequence. ## Pros and Cons at a Glance For readers who want the tradeoffs stated plainly, here are the UniProt cross-reference options for non-model organisms with their pros and condensed cons. **Entry cross-reference section.** Pros are simplicity and curator context. Cons are one-at-a-time workflow and sparse links for non-model species. **ID mapping service.** Pros are bulk speed, official support, and a huge namespace catalog. Cons are silent failures, many-to-many expansion, and no sequence-level verification. **UniParc and PICR.** Pros are proven 100 percent sequence identity across more than 70 source databases, taxonomic and activity filters, and both interactive and programmatic access [1]. Cons are the strict identity bar and the fact that identity is not function. **UniProt BLAST.** Pros are universal applicability and a natural route from raw sequence to full annotation. Cons are similarity rather than identity, paralog traps, and manual verification overhead. **REST API.** Pros are automation and reproducibility at scale. Cons are job management, rate limits, and error handling that lands on you. **Batch downloads.** Pros are total control and permanent reproducibility. Cons are storage cost, maintenance burden, and self-inflicted mapping bugs. ## Common Mistakes and Limitations **Treating a TrEMBL accession as a stable gene identifier.** TrEMBL accessions can change when the underlying nucleotide records change. If your pipeline stores bare TrEMBL accessions as primary keys, version them or key on sequence checksums instead. **Matching identifiers by string pattern.** Guessing that an Ensembl-style prefix or a locus tag format corresponds to RefSeq is not mapping. Use the ID mapping service or sequence identity. String similarity between identifiers from different databases means nothing. **Ignoring isoforms.** A single gene can produce several UniProtKB entries, one canonical and several isoforms. If you map at the gene level, you will silently collapse or expand your protein counts. Decide before the mapping whether your analysis is gene-level or transcript-level. **Forgetting the taxon filter.** Without a taxonomic restriction, sequence and identifier services happily return human or mouse hits for your fish protein. PICR exposes taxonomic identifier filtering as a first-class option precisely to prevent this [1]. Use it every time. **Confusing sequence identity with functional equivalence.** A 100 percent identity match proves the sequences are identical. It does not prove the annotations attached to either record are correct, current, or transferable. This is the deepest limitation of the identity-based approach, and it is a property of the method, not of PICR specifically [1]. Automated annotation in TrEMBL is largely by similarity, so errors propagate through related sequences. **Not recording versions and dates.** A mapping performed today will differ from the same mapping performed after the next RefSeq release. Store the UniProt release, the source database versions, and the job parameters with your results, or your analysis cannot be reproduced. **Not reconciling input and output counts.** Every mapping step should log how many inputs went in and how many rows came out. Expansion and silent drops are normal, but undetected expansion and undetected drops are not. A mapping step that outputs more rows than inputs has duplicated something, and your downstream statistics will inherit the duplication. **Expecting cross-references to be complete or current.** Cross-references are asynchronous with the source databases. A newly added RefSeq record may not appear in UniProt for weeks. For the freshest identifiers, check the source database directly and use UniParc as the tiebreaker. **Believing best BLAST hits are orthologs.** In non-model organisms, the closest sequence in UniProtKB is frequently a paralog or a fast-evolving relative. Confirm orthology with a tree or at minimum with reciprocal best hits before assigning gene names. ## Frequently Asked Questions ### What are UniProt cross-references for non-model organisms? They are links, or maps you build yourself, between protein records for your species in UniProt and the same records in other databases such as RefSeq, Ensembl, and GeneID. For model organisms the links are curated and dense, while for non-model organisms they are sparse and usually must be established by identifier mapping or by exact sequence matching through UniParc [1]. ### What is the best UniProt cross-reference method for a non-model organism? The best method is sequence-based mapping through UniParc or PICR when you need a correctness guarantee, because it requires 100 percent sequence identity across more than 70 source databases and supports taxonomic filtering [1]. The ID mapping service is faster and adequate when your identifiers are recognized and current. Most robust pipelines use both, with identifier mapping first and sequence identity as confirmation. ### How do I use UniProt to convert gene IDs between databases? Use the ID mapping service on uniprot.org or its REST endpoint. Set the source database, the target database, and paste your identifiers, then download the mapped pairs. If the direct conversion is not supported as a single job, map through UniProtKB in two steps and reconcile input and output counts to catch expansions and drops. ### Why do my mapped identifiers stop working after a database update? Because protein identifiers are assigned independently by each database and are frequently retired, merged, or versioned during reannotation. UniParc retains the sequences and the cross-references to source entries even after entries become inactive, and PICR can restrict results to currently active entries so you retrieve only identifiers that resolve today [1]. ### Can I automate UniProt cross-referencing with a script? Yes. The UniProt REST API supports asynchronous ID mapping jobs, entry retrieval, and UniParc queries, and PICR exposes SOAP and REST programmatic interfaces for sequence-based cross-referencing at 100 percent identity [1]. Submit your identifiers or a FASTA file, poll for job completion, stream the results, and log the date and parameters for reproducibility. ### Do identical protein sequences always mean the same function? No. A 100 percent sequence identity match proves the sequences are the same, which is exactly what PICR guarantees, but the annotations attached to either record may be automatic, outdated, or inferred by similarity [1]. Treat identical sequences as strong evidence for functional transfer and verify critical claims against the primary literature. ## Sources 1. [The Protein Identifier Cross-Referencing (PICR) service: reconciling protein identifiers across multiple source databases.](https://pubmed.ncbi.nlm.nih.gov/17945017/)