Datasets:

genomic-benchmarks
/

miRBench

---
license: cc-by-4.0
pretty_name: miRBench (curated)
size_categories:
- 1M<n<10M
tags:
- genomics
- benchmark
- rna
- binary-classification
---
# miRBench, curated

`miRBench`, as published in quality-curated genomic benchmarks - one format, fixed row order, a permanent ID on every row. 3 datasets, 6 files, 2,849,872 rows, one gzipped CSV per split, all Homo sapiens.

## Getting the data

Two packages are the way in: [`genomic-benchmarks-data`](https://pypi.org/project/genomic-benchmarks-data/) for people, [`genomic-benchmarks-data4agents`](https://pypi.org/project/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.

```bash
pip install genomic-benchmarks-data
```

```python
from genomic_benchmarks_data import load_dataset
df = load_dataset("miRBench", "AGO2_CLASH_Hejret2023", "train")
```

A number measured on these files belongs on the [leaderboard](https://genomic-benchmarks.github.io/genomic-benchmarks-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](https://genomic-benchmarks.github.io/) - the project: the QC tool, the curated corpus and the leaderboard.
- [github.com/genomic-benchmarks](https://github.com/genomic-benchmarks) - all three as source, and issues.

## Datasets

| dataset | task | train | test | test_leftout |
|---|---|---|---|---|
| `AGO2_CLASH_Hejret2023` | Binary classification |    8140 |    1018 | - |
| `AGO2_eCLIP_Klimentova2022` | Binary classification | - |     954 | - |
| `AGO2_eCLIP_Manakov2022` | Binary classification | 2495535 |  324171 |   20054 |
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/10.1093/bioinformatics/btaf233
- Covers every dataset here: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/)

## 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/10.1093/bioinformatics/btaf233) 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.

	---
	license: cc-by-4.0
	pretty_name: miRBench (curated)
	size_categories:
	- 1M<n<10M
	tags:
	- genomics
	- benchmark
	- rna
	- binary-classification
	---

	# miRBench, curated

	`miRBench`, as published in quality-curated genomic benchmarks - one format, fixed row order, a permanent ID on every row. 3 datasets, 6 files, 2,849,872 rows, one gzipped CSV per split, all Homo sapiens.

	## Getting the data

	Two packages are the way in: [`genomic-benchmarks-data`](https://pypi.org/project/genomic-benchmarks-data/) for people, [`genomic-benchmarks-data4agents`](https://pypi.org/project/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.

	```bash
	pip install genomic-benchmarks-data
	```

	```python
	from genomic_benchmarks_data import load_dataset

	df = load_dataset("miRBench", "AGO2_CLASH_Hejret2023", "train")
	```

	A number measured on these files belongs on the [leaderboard](https://genomic-benchmarks.github.io/genomic-benchmarks-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](https://genomic-benchmarks.github.io/) - the project: the QC tool, the curated corpus and the leaderboard.
	- [github.com/genomic-benchmarks](https://github.com/genomic-benchmarks) - all three as source, and issues.

	## Datasets

	\| dataset \| task \| train \| test \| test_leftout \|
	\|---\|---\|---\|---\|---\|
	\| `AGO2_CLASH_Hejret2023` \| Binary classification \| 8140 \| 1018 \| - \|
	\| `AGO2_eCLIP_Klimentova2022` \| Binary classification \| - \| 954 \| - \|
	\| `AGO2_eCLIP_Manakov2022` \| Binary classification \| 2495535 \| 324171 \| 20054 \|

	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/10.1093/bioinformatics/btaf233
	- Covers every dataset here: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/)

	## 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/10.1093/bioinformatics/btaf233) 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.