Ensembl Genome Browser

The reference layer under most of modern genomics. Ensembl Genome Browser is the vertebrate genome annotation and variation resource built by EMBL-EBI together with the Wellcome Sanger Institute: every gene, transcript and regulatory feature on a genome, positioned on an assembly (human GRCh38 is the flagship) and stamped with a stable identifier that outlives the release that produced it. A human release carries roughly 60,000 genes, split across protein-coding, lncRNA, miRNA and the rest of the biotype zoo.

Three things make it the join table of the field rather than just another browser. First, stable IDs: ENSG00000141510 is TP53 this release and was last release, so pipelines stop breaking when symbols get renamed. Second, provenance per feature: each record names the annotation pipeline that produced it and the version of its identifier. Third, versioned releases - release 116 is the June 2026 edition, and earlier editions stay addressable, which is what makes a results section reproducible years later.

Get a sample of this dataset and the first thing you will see is real gene records, not screenshots.

One flat record per feature - no envelope, no nesting to unwind. TP53, the most-requested gene in human genetics, exactly as it lands:

id                  : ENSG00000141510
display_name        : TP53
description         : tumor protein p53 [Source:HGNC Symbol;Acc:HGNC:11998]
biotype             : protein_coding
species             : homo_sapiens
assembly_name       : GRCh38
seq_region_name     : 17
start               : 7661779
end                 : 7687546
strand              : -1
canonical_transcript: ENST00000269305.9
logic_name          : ensembl_havana_gene_homo_sapiens
version             : 21

A sequence lookup on the same gene returns the region descriptor with its length:

id             : ENSG00000141510
desc           : chromosome:GRCh38:17:7661779:7687546:-1
molecule       : dna
sequence_length: 25768

Thirteen facts, one row, and every one of them load-bearing: where the gene sits (chromosome 17, 7,661,779-7,687,546, reverse strand), what kind of thing it is (protein_coding), which transcript to treat as canonical, and who annotated it (the joint Ensembl-HAVANA pipeline). Note the description string - it smuggles its own source attribution in brackets, which parsers either strip or exploit.

Fourteen documented fields per gene record, verified against live records during catalog research; thirteen appear below, with full sequence payloads available alongside them. Nothing inferred, no lookup table required to parse a row.

Geography - none, deliberately: these are reference genomes, not survey data. Species coverage spans vertebrates with human GRCh38 at the front, and companion genome collections extend the same identifier scheme to plants, microbes, fungi and invertebrates.

Temporal - annotation arrives as numbered releases, not a moving average. Release 116 is the June 2026 edition; prior releases remain preserved and addressable, which is why "as of release 112" remains a meaningful sentence in a methods section.

Granularity - one record per feature: gene, transcript or variant, each with genomic coordinates. Whole-genome file sets per species sit above that feature level, so a delivery can be a single gene record, a chromosome arm, or everything annotated in one organism.

API, files, or your warehouse. Daily, weekly, or hourly.

• CRISPR and target-design teams - guide design starts with exact coordinates, strand and canonical transcript; a wrong annotation version silently redesigns the experiment. Pinning to a named release removes that failure mode.
• Variant interpretation - a variant only means something once consequence prediction runs against annotated transcripts; stable IDs carry variants from a calling pipeline to clinical and association contexts without symbol drift.
• Cross-database joins - GWAS loci, cancer genomics portals and expression matrices all name genes differently; the ENSG identifier is the common denominator that lets them meet in one table.
• Comparative genomics - orthology and synteny work needs the same annotation conventions across species, which is precisely what a shared identifier scheme buys.
• Reproducibility audits - archived releases make "we re-ran the analysis on the same annotation" a checkable claim instead of a promise.

Data scientists and ML engineers (relevance 3 of 3) pin training corpora to versioned releases so a model's features mean the same thing six months later. Developers and data-product builders (relevance 3) wire annotation into pipelines where latency and bulk both matter, and archived releases give them a rollback path. Journalists, academics and students (relevance 3) cite the canonical EMBL-EBI/Sanger annotation with the release number attached, which is the difference between a citable figure and a screenshot.

Persona packs that lead with trials, patents and market sizing - investors and quants, market researchers, competitive-intel teams - lean on other biotechnology records; this one is infrastructure for anyone whose product touches a genome. See the best biotechnology datasets ranking for where it sits among its peers.

Three quirks worth knowing upfront. First, coordinates are one-based and fully inclusive - position 1 is the first base - so overlap math ported from half-open zero-based tooling needs an off-by-one review before it ships. Second, the description field embeds its attribution in square brackets ("tumor protein p53 [Source:HGNC Symbol;Acc:HGNC:11998]"), a convention that is machine-parseable but trips naive display code. Third, biotype values form an open-ended enum: protein_coding dominates human annotation, but lncRNA, miRNA and artifact classes appear in every release, so treat biotype filters as a whitelist you maintain rather than a closed set you inherit.