GUE_v2, curated

GUE_v2, as published in quality-curated genomic benchmarks - one format, fixed row order, a permanent ID on every row. 32 datasets, 96 files, 846,523 rows, one gzipped CSV per split.

Getting the data

Two packages are the way in: genomic-benchmarks-data for people, genomic-benchmarks-data4agents for agents, the same functions either way. They resolve the URL, check the checksum, and carry each dataset's QC results, which this repository does not.

pip install genomic-benchmarks-data

from genomic_benchmarks_data import load_dataset

df = load_dataset("GUE_v2", "core_promoter_detection_all", "train")

A number measured on these files belongs on the leaderboard, which is what makes two of them comparable, and which prints what QC found about a dataset above every table.

• genomic-benchmarks.github.io - the project: the QC tool, the curated corpus and the leaderboard.
• github.com/genomic-benchmarks - all three as source, and issues.

Datasets

dataset	task	species	train	val	test
core_promoter_detection_all	Binary classification	Homo sapiens	47356	5920	5920
core_promoter_detection_notata	Binary classification	Homo sapiens	42452	5307	5307
core_promoter_detection_tata	Binary classification	Homo sapiens	4904	613	613
enhancer_promoter_interaction_GM12878	Binary classification	Homo sapiens	10000	2000	2000
enhancer_promoter_interaction_HUVEC	Binary classification	Homo sapiens	10000	2000	2000
enhancer_promoter_interaction_HeLa_S3	Binary classification	Homo sapiens	10000	2000	2000
enhancer_promoter_interaction_IMR90	Binary classification	Homo sapiens	10000	2000	2000
enhancer_promoter_interaction_K562	Binary classification	Homo sapiens	10000	2000	2000
enhancer_promoter_interaction_NHEK	Binary classification	Homo sapiens	10000	2000	2000
epigenetic_marks_prediction_H3	Binary classification	Saccharomyces cerevisiae	11971	1497	1497
epigenetic_marks_prediction_H3K14ac	Binary classification	Saccharomyces cerevisiae	26438	3305	3305
epigenetic_marks_prediction_H3K36me3	Binary classification	Saccharomyces cerevisiae	27904	3488	3488
epigenetic_marks_prediction_H3K4me1	Binary classification	Saccharomyces cerevisiae	25341	3168	3168
epigenetic_marks_prediction_H3K4me2	Binary classification	Saccharomyces cerevisiae	24545	3069	3069
epigenetic_marks_prediction_H3K4me3	Binary classification	Saccharomyces cerevisiae	29439	3680	3680
epigenetic_marks_prediction_H3K79me3	Binary classification	Saccharomyces cerevisiae	23069	2884	2884
epigenetic_marks_prediction_H3K9ac	Binary classification	Saccharomyces cerevisiae	22224	2779	2779
epigenetic_marks_prediction_H4	Binary classification	Saccharomyces cerevisiae	11679	1461	1461
epigenetic_marks_prediction_H4ac	Binary classification	Saccharomyces cerevisiae	27275	3410	3410
promoter_detection_300_all	Binary classification	Homo sapiens	47356	5920	5920
promoter_detection_300_notata	Binary classification	Homo sapiens	42452	5307	5307
promoter_detection_300_tata	Binary classification	Homo sapiens	4904	613	613
transcription_factor_prediction_human_0	Binary classification	Homo sapiens	32378	1000	1000
transcription_factor_prediction_human_1	Binary classification	Homo sapiens	30672	1000	1000
transcription_factor_prediction_human_2	Binary classification	Homo sapiens	19000	1000	1000
transcription_factor_prediction_human_3	Binary classification	Homo sapiens	27294	1000	1000
transcription_factor_prediction_human_4	Binary classification	Homo sapiens	19000	1000	1000
transcription_factor_prediction_mouse_0	Binary classification	Mus musculus	6478	810	810
transcription_factor_prediction_mouse_1	Binary classification	Mus musculus	53952	6745	6745
transcription_factor_prediction_mouse_2	Binary classification	Mus musculus	2620	328	328
transcription_factor_prediction_mouse_3	Binary classification	Mus musculus	1904	239	239
transcription_factor_prediction_mouse_4	Binary classification	Mus musculus	15064	1883	1883

Files are v1/<dataset>/<dataset>_<split>.csv.gz, where v1 is the version of the data. A dataset whose sequences change appears under a new prefix rather than being rewritten here.

Licence and provenance

• Collection DOI: https://doi.org/10.48550/arXiv.2306.15006
• Covers every dataset here: Apache-2.0 (https://github.com/MAGICS-LAB/DNABERT_2/blob/main/LICENSE)

Every row has an identity

Each file carries two columns the upstream release does not:

• id - <collection>:<dataset>:<version>:<split>:<row index>, the index zero-padded to 7 digits.
• row_index - the row's 0-based position in this file.

Row order is fixed within a major version, so an id names the same sequence for as long as that version exists - which is what makes a reported number traceable to what it was computed on. The corpus metadata carries the SHA-256 of every split file, so a copy can be checked against the audited one.

Citation

Cite the collection's own publication (https://doi.org/10.48550/arXiv.2306.15006) for the data, and Automated quality control for genomic sequence benchmarks reveals pervasive single-feature separability and train-test leakage for the curation and the audit.