# Ensembl Gene vs. Transcript Sequences: What's the Difference In Ensembl, a gene is an annotated genomic region that can produce one or more RNA products, and its "sequence" is the full stretch of genomic DNA between the gene's start and end coordinates, introns included. A transcript is a single spliced RNA product of that gene, and its "sequence" is the stitched-together exons of that one product, usually presented as cDNA, coding sequence (CDS), or protein. This distinction sounds academic until the day it changes your results. Choose the gene sequence when you need the transcript, and your BLAST hits get riddled with intron gaps. Choose a single transcript when the biology depends on isoforms, and you may miss the variant, exon, or probe set that carries the signal. Because Ensembl annotates tens of thousands of human transcripts across tens of thousands of genes, and because the annotation itself changes every release [1], knowing exactly which sequence you pulled, and why, is one of the most practical skills in genomics. Ensembl itself now covers more than 4,800 eukaryotic and more than 31,300 prokaryotic genomes [2], so this gene-versus-transcript logic applies far beyond human data. This article explains what each sequence is, how to retrieve both, and how the choice between them affects variant interpretation, expression analysis, and homology work. ## How Ensembl defines a gene and a transcript ### The gene record: a genomic address with coordinates An Ensembl gene is fundamentally a coordinate object. It has a stable identifier (for human genes, an accession starting with ENSG, such as ENSG00000139618 for BRCA2), a chromosome, a start position, an end position, a strand, and a biotype such as protein_coding, lincRNA, or pseudogene. The gene does not have one canonical sequence of its own. The "gene sequence" you download is simply the genomic DNA spanning those coordinates, on the correct strand. That span includes everything between the outermost boundaries: - All exons, coding and noncoding - All introns, which are spliced out of mature RNA but remain in the genomic sequence - Untranslated regions (UTRs), the parts of exons that flank the coding portion - Any overlapping regulatory or noncoding features that happen to fall inside the span For most human protein-coding genes, this span is tens of kilobases, and some genes stretch well beyond a million bases. When you ask for a "gene sequence," you are asking for a piece of chromosome. ### The transcript record: a spliced product An Ensembl transcript has its own identifier (ENST for human, such as ENST00000381657 for a widely used BRCA2 transcript), its own exon structure, and its own set of derived sequences. The transcript model tells the annotation pipeline which exons exist, in what order, and which stretch of that spliced product is coding. Ensembl offers four sequence views for every transcript: 1. **cDNA**: the mature, spliced RNA sequence, exons only, with UTRs included 2. **CDS**: the coding portion only, from the first ATG to the stop codon 3. **Peptide**: the translated [protein sequence](/blog/guides/protein-sequence) that the CDS would produce 4. **Genomic sequence**: the unspliced genomic DNA that this transcript spans, which is a subset of the gene span The cDNA is what most people mean when they say "transcript sequence." It corresponds to what a reverse-transcribed mature mRNA would look like, and it is the reference against which RNA-seq reads are typically aligned and against which variant effects are usually judged. ### One gene, many transcripts The core of the ensembl gene sequence vs transcript sequence question is [alternative splicing](/blog/guides/alternative-splicing). A single gene can produce multiple transcripts with different exon combinations, different promoters, and different 3' ends. Ensembl annotates all credible isoforms it has evidence for, so the number of transcripts per gene varies enormously. Some genes have one well-supported transcript. Others have dozens. In the 2023 human annotation, the lncRNA gene PCBP1-AS1 carried 296 annotated isoforms, the most of any gene in that release [1]. Because each transcript is a separate record with a separate sequence, a "gene sequence" cannot tell you which isoform was measured, which exon was skipped, or which protein variant was produced. Only the transcript level carries that information. ## What each sequence actually contains ![Decision path asking whether the work is about a locus, an expressed isoform, or a protein, and naming the sequence type to download for each.](/article-images/illustrations/ensembl-gene-sequence-vs-transcript-sequence-sequence-choice.webp) The following table summarizes the object types you will encounter when using Ensembl, and how they relate to one another. | Object | Identifier prefix (human) | What the sequence contains | Typical use | |--|--|--|--| | Gene (genomic) | ENSG | Full genomic span, exons plus introns plus UTRs, unspliced | Locating a locus, designing intronic assays, promoter and regulatory searches | | Transcript (cDNA) | ENST | Spliced exons in transcript order, UTRs included | RNA-seq references, isoform studies, transcript BLAST | | CDS | ENST (CDS view) | Coding bases only, from start codon to stop codon, no UTRs | Codon-level analysis, ORF checks | | Protein (peptide) | ENSP | Translated [amino acid sequence](/blog/guides/amino-acid-sequence) of one CDS | Protein BLAST, domain and structure work | | Exon | ENSE | One individual exon's genomic sequence | Exon-level probes, junction analysis | Two relationships are worth memorizing. First, the transcript sequence is always shorter than or equal to the gene sequence, because introns are removed. A gene span of 84 kilobases can yield a mature transcript of a few kilobases. Second, the transcript is always at least as long as the CDS, because UTRs flank the coding region, and 5' and 3' UTRs are frequently longer than the coding sequence itself. A third distinction causes frequent confusion: Ensembl's annotation comes in tiers. For human and mouse, the Ensembl annotation is produced jointly with the GENCODE project, and the GENCODE Comprehensive set is demonstrably richer in alternative splicing, novel coding sequences, novel exons, and overall genomic coverage than RefSeq, while the GENCODE Basic set, a reduced set of representative transcripts per gene, closely resembles RefSeq [3]. When you download sequences, the tag column in the GTF tells you whether a transcript belongs to the basic set or only to the comprehensive set. ## Ensembl gene sequence vs transcript sequence explained with a worked example Consider BRCA2, the breast cancer susceptibility gene on chromosome 13. The Ensembl gene record spans roughly 84 kilobases of genomic DNA organized into about 27 exons. If you download the gene sequence, you get those 84 kilobases with every intron in place, and the coding exons account for only a modest fraction of that span. If instead you open one well-supported BRCA2 transcript and download its cDNA, you get the spliced exons, roughly the length of the mature mRNA, with long 5' and 3' UTRs. Download the CDS and you get just the coding frame. Download the peptide and you get the 3,418-amino-acid BRCA2 protein. All four sequences derive from the same locus, yet they answer completely different questions: - Screening for a deep intronic variant that affects splicing? You need the genomic (gene) sequence. - Aligning RNA-seq reads to ask which isoform is expressed? You need the transcript cDNAs. - Annotating a missense mutation found in a patient? You need the specific transcript's CDS and protein, because residue numbering depends on the transcript chosen. - Searching for homologs of the protein? You need the peptide. This worked comparison is the fastest way to understand what the ensembl gene sequence vs transcript sequence distinction means in practice. The same locus, four biologically distinct sequences. ## How to use Ensembl to retrieve each sequence Learning how to use Ensembl is mostly learning where the sequence export options live. The Ensembl browser provides a comprehensive, integrated source of genome annotation with links to external resources, and navigating to a gene of interest is the standard entry point for most users [4]. Ensembl is also updated on a regular release cycle (roughly quarterly), and downstream tools such as g:Profiler synchronize their own annotation with Ensembl releases on that same quarterly schedule [5], which is why recording the release number matters for reproducibility. ### Through the genome browser A quick ensembl [genome browser tutorial](/blog/guides/genome-browsers-for-researchers-a-guide-to-inspecting-genomic-evidence), in four steps: 1. Go to ensembl.org, type your gene name or identifier into the search box, and select the species and assembly (for human, GRCh38). 2. On the gene page, check the Location tab for coordinates, strand, and biotype. This tab defines what "the gene sequence" means for this record. 3. Open the Transcripts table. Each row is one isoform, color-coded by biotype, with flags such as MANE Select, Ensembl canonical, and RefSeq match. Click a transcript to open its transcript page. 4. On the transcript page, use "Export data" to download cDNA, CDS, peptide, or genomic sequence. The gene page's "Export data" option gives you the full unspliced genomic span. The MANE Select flag deserves attention. MANE (Matched Annotation from NCBI and EMBL-EBI) is a joint effort between Ensembl/GENCODE and RefSeq to define a standard transcript per gene. Each MANE transcript is an exact match in exonic sequence between the Ensembl and RefSeq records, so identifiers can be used synonymously across resources, and MANE Select transcripts now cover 97 percent of human protein-coding genes, including every gene on the American College of Medical Genetics and Genomics Secondary Findings list v3.0 [6]. If you need one defensible transcript per gene, MANE Select is the default answer. ### Through BioMart and bulk downloads For hundreds or thousands of sequences, the browser becomes tedious. Two bulk routes exist. **BioMart** (reachable from the Ensembl home page) lets you build a query such as "give me all Ensembl transcript IDs, gene symbols, biotypes, and cDNA sequences for protein_coding transcripts on chromosome 7." You choose filters (species, region, biotype), choose attributes (IDs, sequences), and export a table. BioMart is the fastest point-and-click route to sequence lists. **FTP and file formats** are the reproducible route. Every release publishes FASTA files for unspliced genomic DNA, cDNA, CDS, and peptides, plus GTF and GFF3 files that define every gene and transcript model. For example: ``` wget https://ftp.ensembl.org/pub/release-113/gtf/homo_sapiens/Homo_sapiens.GRCh38.113.gtf.gz wget https://ftp.ensembl.org/pub/release-113/fasta/homo_sapiens/cdna/Homo_sapiens.GRCh38.cdna.all.fa.gz ``` The GTF file is what aligners such as STAR and HISAT2 consume, and it encodes the gene-versus-transcript hierarchy directly: each line has a feature type (gene, transcript, exon) with parent-child relationships via gene_id and transcript_id attributes. Building a transcriptome index for RNA-seq from the cdna FASTA and a genome index from the DNA FASTA is the most common pairing of the two sequence types in practice. Plant researchers use the same interfaces through Ensembl Plants, which provides genome sequence, gene models, and functional annotation for dozens of plant species and updates four to six times per year [7]. ### Through the REST API For scripting, the Ensembl REST API returns sequence by identifier. To fetch the unspliced genomic sequence of a gene: ``` curl "https://rest.ensembl.org/sequence/id/ENSG00000139618?content-type=text/x-fasta&type=genomic" ``` To fetch the spliced cDNA of a transcript: ``` curl "https://rest.ensembl.org/sequence/id/ENST00000381657?content-type=text/x-fasta&type=cdna" ``` Changing the type parameter to cds or protein switches the returned product. The type parameter is exactly where the gene-versus-transcript distinction becomes code: a genomic type request against a transcript ID returns that transcript's unspliced genomic span, while a cdna request against a gene ID returns an error, because a gene has no single cDNA. ### Through the MySQL server and programmatic access Ensembl also runs a public read-only MySQL server (ensembldb.ensembl.org on port 5306, user anonymous) where the gene, transcript, exon, and translation tables make the hierarchy explicit. Most users will prefer the API libraries (Perl, Python, or R through biomaRt), which wrap the same relational structure. ## Why the choice changes your results ![Comparison of GENCODE Comprehensive, GENCODE Basic, RefSeq and MANE Select on splicing richness and whether Ensembl and RefSeq records match exactly.](/article-images/illustrations/ensembl-gene-sequence-vs-transcript-sequence-annotation-sets.webp) ### Variant interpretation is transcript-dependent This is where the ensembl gene sequence vs transcript sequence interpretation question stops being theoretical. A variant's consequence, missense versus synonymous, benign versus loss-of-function, depends entirely on which transcript you project it onto. A variant in an exon shared by all isoforms has one meaning. A variant in an exon present in only some isoforms has a transcript-limited meaning. A variant in the 3' UTR of one isoform may fall inside the coding sequence of another. The scale of this effect is documented. A systematic comparison of GENCODE and RefSeq annotation found that the choice of reference transcript set produces large differences in variant annotation, with loss-of-function consequences discordant for at most about 30 percent of LoF variants, with the disagreement concentrated in noncoding transcripts and UTR sequence rather than well-annotated coding exons [3]. Tools built on Ensembl data inherit this transcript dependence: the Ensembl Variant Effect Predictor (VEP), which is under active expansion in current Ensembl development [2], annotates variants against specific transcripts, and published clinical workflows explicitly report variants against a named transcript accession, as in a recent study that annotated a ChAT missense variant as c.1213C>G on transcript NM_020549.5 using VEP [8]. Changing the transcript changes the c. numbering, the protein residue number, and sometimes the predicted consequence. The clinical answer to this ambiguity is MANE, described above, which fixes one agreed transcript per protein-coding gene so that variant reports mean the same thing everywhere [6]. ### Isoform-level biology needs transcript sequences, not gene sequences Modern variant and protein analyses operate at the isoform level. One computational study of esophageal cancer genes identified 126 protein isoforms across 28 genes from NCBI and Ensembl records and mapped ten rare coding variants onto 23 transcript-level variants, showing that the same gene-level SNP can have different structural consequences depending on the isoform it lands in [9]. Resources such as HapScoreDB push this further, scoring phased protein haplotypes across all human transcript isoforms by combining GENCODE and Ensembl annotations with phased 1000 Genomes variants, yielding more than 130,000 distinct protein haplotypes from over 18,000 genes and 78,000 transcripts [10]. None of that is possible from gene sequences alone. Transcript-level structure is also becoming a [comparative genomics](/blog/guides/comparative-genomics) object in its own right. TranscriptDB catalogs conserved transcripts, genes whose homologs retain similar exon composition, across 317 annotated eukaryotic species drawn from Ensembl release 111, enabling transcript phylogenies alongside gene phylogenies [11]. ### Expression analysis: probes, genes, and isoforms Expression platforms force the same decision. Microarray probes are short oligonucleotides that must be mapped to transcripts before results can be interpreted, and how you group transcripts into genes changes the answer. In a zebrafish microarray study, remapping probes against an integrated annotation built from RefSeq, GenBank, and Ensembl transcripts retained up to 20 percent more probes and detected roughly 1,000 more genes than remapping against any single database alone [12]. The lesson generalizes: gene-level summarization hides isoform changes, and transcript-level summarization requires transcript sequences as the reference. RNA-seq has the same fork. Align to the genome with a splice-aware aligner plus the GTF (gene-and-transcript model as coordinates), or pseudoalign to the cDNA FASTA (transcript sequences directly). Quantification tools then report at either level, and gene-level counts in tools such as those synchronized with Ensembl annotation are the input for downstream functional analysis, with g:Profiler alone supporting 213 species for enrichment and identifier mapping [5]. ### Homology and cross-species mapping Most cross-species gene mapping still starts from Ensembl homology tables, which connect genes across species. One recent benchmark in cross-species single-cell integration noted that the default many-to-many Ensembl homology mappings can overweight expanded gene families, and that enforcing one-to-one mapping reduces the number of homologous pairs by roughly 8 percent, about 900 pairs in their benchmark [13]. The gene-level unit dominates this workflow because orthology is usually defined for genes, not isoforms, which is a legitimate reason to work at the gene level. The reverse trap is assuming gene-level homology implies transcript-level equivalence, since corresponding genes in two species often have non-corresponding isoform repertoires [11]. ### Annotation drift and versioning Ensembl annotations are not frozen. A longitudinal characterization of human Ensembl annotations from 2014 to 2023 documented growth from 58,812 genes in 2014 to 62,710 in 2023, with some releases showing dips, and with the protein-coding gene count rising from 19,953 in 2014 to 20,048 in 2023 [1]. The same study demonstrated the biological consequence of this drift by quantifying expression of isoforms newly annotated between 2019 and 2023 across nine GTEx tissues, finding 2,054 of these newly annotated isoforms expressed in cerebellar hemisphere and 594 in liver, often at relative levels exceeding the previously known isoforms [1]. A transcript sequence you downloaded in 2019 may not exist in the 2024 release, and a gene's transcript count may have changed under the same gene ID. ### Noncoding genes and regulatory regions The gene-versus-transcript distinction behaves differently outside protein-coding genes. Noncoding RNA genes often have transcripts with little or no splicing, and gene-centric resources for ncRNAs exist precisely because so many ncRNA databases are transcript-centric. GeneCaRNA, for instance, integrates Ensembl, HGNC, NCBI Gene, and RNAcentral into 219,587 genome-mapped human ncRNA gene pages, a sevenfold increase over the gene-mining sources it draws from [14]. For regulatory annotation, the gene boundary is only a rough guide, since promoters and enhancers frequently sit outside the annotated gene span, and promoter prediction studies explicitly compare results across Ensembl, RefSeq, and other annotation systems, as done for the rice genome with three annotations including Ensembl [15]. ## Common mistakes and limitations **Downloading the gene sequence when the transcript sequence is required.** This is the single most common error. If a downstream tool expects cDNA (BLAST against an nr transcript database, primer design on the mRNA, ORF prediction), an unspliced genomic sequence produces gaps, intronic hits, and nonsense ORFs. Check whether your input is genomic or spliced before running anything. **Ignoring the version number after the ID.** An accession such as ENST00000381657 without a version is ambiguous across releases. The version suffix increments when the model changes. Record the full versioned accession and the Ensembl release number in every methods section. **Mixing annotation releases within one analysis.** Because annotation content shifts between releases, with gene and isoform counts changing year to year [1], combining a GTF from one release with FASTA files from another causes transcript IDs to mismatch and quantification to fail silently. Pull all files for one analysis from the same release directory. **Assuming "canonical" means "clinically correct."** Ensembl flags one transcript as canonical per gene, and RefSeq has its own convention. These do not always agree with each other. When consistency across resources matters, prefer MANE Select, whose Ensembl and RefSeq records are exonic-sequence-identical by construction [6]. **Treating UTRs and 3' ends as complete.** Annotation of gene 3' ends remains incomplete even in major databases. Comparative analysis across vertebrates has identified hundreds of unannotated 3' UTR extensions, including several hundred novel extensions in rat, and well-studied human loci such as ADD2 and CPEB3 had conserved alternative 3' isoforms missing from both Ensembl and RefSeq at the time of that study [16]. A transcript sequence is an annotation, not a measurement, and its UTR boundaries can change. **Confusing biotype with confidence.** A transcript flagged protein_coding is not automatically well supported, and a lncRNA annotation is not automatically spurious. Look at evidence tags and TSL (transcript support level) values in the GTF rather than inferring quality from biotype. **Forgetting species prefixes in identifiers.** Human accessions begin with ENS followed directly by the object letter, but other species carry prefixes such as ENSMUS for mouse. A script that validates IDs against one pattern will reject valid records from other species, and Ensembl's multi-species scale, now more than 4,800 eukaryotic genomes [2], means mixed-prefix errors are common in large collations. **Reading too much into gene-level homology.** Ensembl homology tables are gene-centric, and relying on default many-to-many mappings can distort downstream analyses [13]. If your question is about a specific protein product, verify that the homologous transcript exists and is expressed in the species you are comparing. **Assuming the gene sequence contains the regulatory picture.** Promoters, enhancers, and boundary elements often lie outside the annotated gene span, and their identification depends on which annotation system and evidence tracks you trust, as rice promoter-annotation comparisons showed when results differed across Ensembl, RefSeq, and project-specific annotations [15]. One general limitation applies to every workflow described here: individual biological conclusions, especially clinical ones, require expert review and validation beyond any single annotation resource. ## Quick Review - A gene in Ensembl is a coordinate span on the genome. Its sequence is unspliced genomic DNA including introns. - A transcript in Ensembl is one spliced product. Its sequence is cDNA (exons plus UTRs), with CDS and peptide derived from it. - One gene can have many transcripts, each with its own ENST identifier, exon structure, and sequence. - Variant annotation, c. numbering, and protein predictions are all transcript-specific, and reference transcript choice changes variant interpretation measurably [3]. - MANE Select transcripts, available for 97 percent of human protein-coding genes, provide an exonic-sequence-identical standard shared between Ensembl and RefSeq [6]. - Annotation content changes every release, so record the Ensembl release number and full versioned accessions [1]. - Match the sequence type to the question: genomic for loci and splicing context, cDNA for expression, CDS and peptide for coding and protein work. ## Frequently Asked Questions ### Can I use a gene sequence as a BLAST query against a transcript database? You can, but the results will be noisy. The introns in your query have no counterparts in transcript databases, so alignments break into exonic fragments separated by large gaps. For transcript-level searches, use the cDNA. For searching a genome, use the genomic sequence against a genomic database. ### Which transcript should I choose when a gene has many? Default to MANE Select when the species is human, because it is an exact exonic match between Ensembl/GENCODE and RefSeq and exists for 97 percent of protein-coding genes [6]. Outside human data, use the Ensembl canonical transcript, and state in your methods that you did so. If the biology is isoform-specific, analyze all transcripts rather than one. ### What is the difference between cDNA, CDS, and peptide sequences in Ensembl? The cDNA is the full spliced transcript including UTRs. The CDS is the coding portion from the start codon to the stop codon, with UTRs removed. The peptide is the protein sequence translated from the CDS. All three are available from the same transcript record through the export menu or the REST API. ### Why does my Ensembl transcript ID end with a number after a dot? The suffix is a version number that increments whenever the transcript model changes between releases. Versioned accessions let you reproduce results against an exact annotation state. Citing a bare accession without the version leaves the model ambiguous. ### Do Ensembl transcripts match RefSeq transcripts? Not in general. The GENCODE Comprehensive set that backs Ensembl human annotation is richer in alternative splicing, novel exons, and coding sequence than RefSeq, while the reduced GENCODE Basic set closely resembles RefSeq [3]. The exception is the MANE set, where each transcript is an exact exonic match between the two resources so identifiers can be used synonymously [6]. ### How do I download all gene and transcript sequences for a genome? Use the Ensembl FTP site. For one analysis, download the genome DNA FASTA, the cDNA FASTA, and the GTF annotation file from the same release directory, for example release-113 for human. The GTF defines gene and transcript models as coordinates, the DNA FASTA holds genomic sequence, and the cDNA FASTA holds spliced transcript sequences ready for expression quantification. ## Sources 1. [Genome annotations matter: characterizing Ensembl hg38 annotations from 2014 to 2023.](https://pubmed.ncbi.nlm.nih.gov/41345834/) 2. [Ensembl 2025.](https://pubmed.ncbi.nlm.nih.gov/39656687/) 3. [Comparison of GENCODE and RefSeq gene annotation and the impact of reference geneset on variant effect prediction.](https://pubmed.ncbi.nlm.nih.gov/26110515/) 4. [Using the Ensembl genome server to browse genomic sequence data.](https://pubmed.ncbi.nlm.nih.gov/18428779/) 5. [g:Profiler-a web server for functional interpretation of gene lists (2016 update).](https://pubmed.ncbi.nlm.nih.gov/27098042/) 6. [A joint NCBI and EMBL-EBI transcript set for clinical genomics and research.](https://pubmed.ncbi.nlm.nih.gov/35388217/) 7. [Ensembl Plants: Integrating Tools for Visualizing, Mining, and Analyzing Plant Genomic Data.](https://pubmed.ncbi.nlm.nih.gov/27987162/) 8. [The identification of novel missense variant in ChAT gene in a patient with gestational diabetes denotes plausible genetic association.](https://pubmed.ncbi.nlm.nih.gov/40688037/) 9. [Computational identification of rare pathogenic genomic variants in esophageal cancer markers: Transcript-level analysis, sequence-based insights, and structural-functional impacts of non-synonymous SNPs.](https://pubmed.ncbi.nlm.nih.gov/41777455/) 10. [HapScoreDB: a database of protein language model functional scores for haplotype-resolved protein sequences.](https://pubmed.ncbi.nlm.nih.gov/41261743/) 11. [TranscriptDB: a transcript-centric database to study eukaryotic transcript conservation and evolution.](https://pubmed.ncbi.nlm.nih.gov/39530236/) 12. [Integrating multiple genome annotation databases improves the interpretation of microarray gene expression data.](https://pubmed.ncbi.nlm.nih.gov/20089164/) 13. [Protein large language model-assisted one-to-one gene homology mapping in cross-species single-cell transcriptome integration.](https://pubmed.ncbi.nlm.nih.gov/42855280/) 14. [GeneCaRNA: A Comprehensive Gene-centric Database of Human Non-coding RNAs in the GeneCards Suite.](https://pubmed.ncbi.nlm.nih.gov/33676929/) 15. [Effect of a Comparative Analysis of Different Annotations of the Oryza sativa Rice Genome for In Silico Verification of Predicted Promoter Sequences.](https://pubmed.ncbi.nlm.nih.gov/40976834/) 16. [Identification of 3' gene ends using transcriptional and genomic conservation across vertebrates.](https://pubmed.ncbi.nlm.nih.gov/23244605/)