altar-genomics/opentargets-e2g-26.09

Open Targets enhancer-to-gene links 26.09, GRCh38

ENCODE-rE2G element-to-gene links from the Open Targets Platform release 26.09 (2026-09-24) enhancer_to_gene dataset, in the Parquet link layout that the Altar altar-opentargets-e2g binding reads: 48,806,245 rows in 23 files (2.7 GB).

This is a derived copy published by the Altar maintainers, not an Open Targets product. The authoritative data is Open Targets' release at https://ftp.ebi.ac.uk/pub/databases/opentargets/platform/26.09/output/enhancer_to_gene/.

What it holds

Each row is one predicted regulatory link: an element (a 0-based, half-open interval on GRCh38) and a gene it is predicted to regulate, in one ENCODE biosample, scored by ENCODE-rE2G (Gschwind et al. 2023). Open Targets assembled 1,439 ENCODE-rE2G prediction files (322 biosamples, 20,088 genes) and keeps links scored at or above 0.6; so does this copy. A variant overlapping an element is linked to that element's genes.

How it was made

altar-opentargets-e2g-build (builder version 0.3.0rc1) read Open Targets' 30 enhancer_to_gene Parquet files (48,808,491 rows), mapped each row through the binding's OpenTargetsE2GRow, the same mapping its link-store importer uses, and wrote the rows with Altar's write_link_dataset. 2,246 rows repeated another row's intervalId; the builder checked that each repeat was identical to the first, column for column, and kept one. Every file's footer records altar_links.release=26.09, altar_links.genome_build=hg38, altar_links.dataset_namespace=org.opentargets.platform, altar_links.dataset_name=enhancer-to-gene, altar_links.dataset_version=1 and altar_links.layout=variant-gene-links/1.

Layout and columns

Files are Hive-partitioned by chromosome (chromosome=1/links.parquet, …, chromosome=X/links.parquet; no chr prefix), and chromosome is also stored in every file. Rows are sorted by start, in row groups of 32,000 rows, so an overlap query on start and end reads only the row groups it can match.

Column	Meaning
chromosome, start, end	The element: 1–22 or X, 0-based, half-open, GRCh38
genome_build	GRCh38
gene_id, gene_namespace	The target gene's Ensembl ID (ENSG…) and ensembl_gene
gene_symbol	Empty: Open Targets gives none
score, score_direction	The ENCODE-rE2G score (≥ 0.6) and higher_is_stronger
biosample, biosample_id	The biosample's name and ontology ID (CL:, UBERON:, EFO:)
regulatory_element_id, element_type	GRCh38:<chromosome>:<start>-<end>, and Open Targets' intervalType (promoter, genic, intergenic)
source_id, source_name, method	E2G, Open Targets Platform, ENCODE-rE2G
data_release	26.09
record_id	opentargets:26.09:<intervalId>, unique
source_record_id, source_dataset	Open Targets' intervalId and enhancer_to_gene

Open Targets' component scores, distance to the TSS, PubMed ID, quality controls and studyId are not columns; source_record_id finds the row in Open Targets' release, which holds them.

Use

With Altar, the opentargets_e2g source reads the published copy by default and adds per-variant link columns (link_top_gene, link_top_score, link_count, link_genes, …):

[sources.e2g]
type = "opentargets_e2g"
options = { location = "hf" }  # needs altar[hf]; the default reads gs://

Hugging Face limits anonymous requests, so a large cohort read from this mirror can stall. For a large cohort, read the gs:// copy, the default, or copy the files once. altar pull opentargets_e2g --to <path-or-uri> copies it into storage you own, checking each file's release.

Without Altar, any Parquet reader works. With DuckDB, the links overlapping a variant at chr7:140,924,700 (1-based), in BRAF's promoter:

SELECT gene_id, score, biosample, element_type
FROM 'hf://datasets/altar-genomics/opentargets-e2g-26.09/chromosome=7/*.parquet'
WHERE start < 140924700 AND "end" > 140924699
ORDER BY score DESC;

The same files are published at gs://kundajelab-altar/opentargets-e2g/enhancer-to-gene-26.09/, readable anonymously.

Terms and citation

Open Targets Platform data is released under CC0 1.0, and so is this copy. Open Targets asks users to cite its latest publication, and the predictions come from ENCODE-rE2G and ENCODE data; please cite:

• Buniello A, et al. Open Targets Platform: facilitating therapeutic hypotheses building in drug discovery. Nucleic Acids Research 2025;53(D1):D1467–D1475. doi:10.1093/nar/gkae1128.
• Gschwind AR, Mualim KS, Karbalayghareh A, et al. An encyclopedia of enhancer-gene regulatory interactions in the human genome. bioRxiv 2023. doi:10.1101/2023.11.09.563812. PMID 38014075.
• The ENCODE Project Consortium. Expanded encyclopaedias of DNA elements in the human and mouse genomes. Nature 2020;583:699–710. doi:10.1038/s41586-020-2493-4. ENCODE asks that you also cite the accessions of the files you use: the ENCODE-rE2G prediction files are ENCFF… accessions under ENCODE experiments (ENCSR…), and Open Targets names each row's file in its studyId column.