README.md

---
license: cc-by-4.0
task_categories:
  - tabular-regression
tags:
  - biology
  - oligonucleotides
  - siRNA
  - ASO
  - shRNA
  - HELM
  - oligogym
configs:
  - config_name: ichihara_2007_1
    data_files:
      - split: train
        path: ichihara_2007_1.csv.gz
  - config_name: ichihara_2007_2
    data_files:
      - split: train
        path: ichihara_2007_2.csv.gz
  - config_name: alharbi_2020_1
    data_files:
      - split: train
        path: alharbi_2020_1.csv.gz
  - config_name: alharbi_2020_2
    data_files:
      - split: train
        path: alharbi_2020_2.csv.gz
  - config_name: hagedorn_2022_1
    data_files:
      - split: train
        path: hagedorn_2022_1.csv.gz
  - config_name: hwang_2024_1
    data_files:
      - split: train
        path: hwang_2024_1.csv.gz
  - config_name: knott_2014_1
    data_files:
      - split: train
        path: knott_2014_1.csv.gz
  - config_name: moe_neurotox_1
    data_files:
      - split: train
        path: moe_neurotox_1.csv.gz
  - config_name: martinelli_2023_1
    data_files:
      - split: train
        path: martinelli_2023_1.csv.gz
  - config_name: mcquisten_2007_1
    data_files:
      - split: train
        path: mcquisten_2007_1.csv.gz
  - config_name: papargyri_2020_1
    data_files:
      - split: train
        path: papargyri_2020_1.csv.gz
  - config_name: shmushkovich_2018_1
    data_files:
      - split: train
        path: shmushkovich_2018_1.csv.gz

---

# OligoGym Datasets

A mirror of the 12 curated benchmark datasets shipped with
[**OligoGym**](https://github.com/Roche/OligoGym) (Roche), repackaged as HuggingFace configs for
convenient loading. Covers oligonucleotide activity, toxicity, and
immunomodulation across ASO, siRNA, and shRNA modalities.

> **This is not a new dataset.** Row counts, column sets, and values are
> identical to the files in OligoGym's `oligogym/resources/pkg_dataset/`.
> All credit for curation belongs to the OligoGym authors — please cite them
> (see [Citation](#citation)) **and** each dataset's own primary source.

## Columns

| Column | Description |
|---|---|
| `x` | HELM string for the oligonucleotide |
| `y` | Numeric label, as processed by OligoGym (the modelling target) |
| `y_raw` | Original label before clipping / scaling — differs from `y` for `hwang_2024_1`, `knott_2014_1`, `moe_neurotox_1`, `papargyri_2020_1` |
| `targets` | mRNA target identifier |
| `smiles` | SMILES representation |
| `fasta` | Nucleotide sequence |

`hwang_2024_1` additionally carries `cell_line` and `concentration`.

## Datasets

| Key | Canonical Name | OligoGym ID | Modality | Collection | Samples | Source(s) |
|---|---|---|---|---|---|---|
| `ichihara_2007_1` | Ichihara_2007_1 | Huesken | SIRNA | activity | 2,431 | [DOI](https://doi.org/10.1093/nar/gkm699) |
| `ichihara_2007_2` | Ichihara_2007_2 | Ichihara | SIRNA | activity | 419 | [DOI](https://doi.org/10.1093/nar/gkm699) |
| `alharbi_2020_1` | Alharbi_2020_1 | TLR7 | ASO | immunomodulation | 192 | [DOI](https://doi.org/10.1093/nar/gkaa523) |
| `alharbi_2020_2` | Alharbi_2020_2 | TLR8 | ASO | immunomodulation | 192 | [DOI](https://doi.org/10.1093/nar/gkaa523) |
| `hagedorn_2022_1` | Hagedorn_2022_1 | Neurotox LNA | ASO | toxicity | 1,825 | [DOI](https://doi.org/10.1089/nat.2021.0071) |
| `hwang_2024_1` | Hwang_2024_1 | ASOptimizer | ASO | activity | 32,602 | [DOI](https://doi.org/10.1016/j.omtn.2024.102186) |
| `knott_2014_1` | Knott_2014_1 | Sherwood | SHRNA | activity | 291,551 | [DOI](https://doi.org/10.1016/j.molcel.2014.10.025) |
| `moe_neurotox_1` | MOE_Neurotox_1 | Neurotox MOE | ASO | acute neurotoxicity | 2,437 | [1](https://patents.google.com/patent/WO2020172559A1) [2](https://patents.google.com/patent/WO2016168592A2) [3](https://patents.google.com/patent/WO2021021673A1) [4](https://patents.google.com/patent/WO2015153800A2) [5](https://patents.google.com/patent/WO2020205463A1) [6](https://patents.google.com/patent/WO2020006267A1) [7](https://patents.google.com/patent/WO2020023737A1) [8](https://patents.google.com/patent/WO2020243292A1) [9](https://patents.google.com/patent/WO2020106996A1) [10](https://patents.google.com/patent/WO2022026589A1) [11](https://patents.google.com/patent/WO2022165122A1) [12](https://patents.google.com/patent/WO2021174019A1) [13](https://patents.google.com/patent/WO2022032060A2) |
| `martinelli_2023_1` | Martinelli_2023_1 | siRNAmod | SIRNA | activity | 907 | [DOI](https://doi.org/10.1016/j.ygeno.2024.110815) |
| `mcquisten_2007_1` | McQuisten_2007_1 | OpenASO | ASO | activity | 3,913 | [1](https://doi.org/10.1186/1471-2105-8-184) [2](https://github.com/lackeylela/openASO) |
| `papargyri_2020_1` | Papargyri_2020_1 | Cytotox LNA | ASO | toxicity | 768 | [DOI](https://doi.org/10.1016/j.omtn.2019.12.011) |
| `shmushkovich_2018_1` | Shmushkovich_2018_1 | Shmushkovich | SIRNA | activity | 356 | [DOI](https://doi.org/10.1093/nar/gky745) |

`moe_neurotox_1` is compiled from 13 separate WO (PCT) patent publications;
`mcquisten_2007_1` from a publication plus the openASO repository. All sources
are linked above.

## Usage

```python
from datasets import load_dataset

# Load a single dataset
ds = load_dataset("CollageBio/oligogym", name="ichihara_2007_1")
df = ds["train"].to_pandas()

# Or download the raw file directly
# https://huggingface.co/datasets/CollageBio/oligogym/resolve/main/ichihara_2007_1.csv.gz
```

## Citation

If you use these datasets, please cite the OligoGym paper:

```bibtex
@article{rotrattanadumrong2026oligogym,
  title={OligoGym: Curated Datasets and Benchmarks for Oligonucleotide Drug Discovery},
  author={Rotrattanadumrong, Rachapun and De Donno, Carlo},
  journal={Advances in Neural Information Processing Systems},
  volume={38},
  year={2026}
}
```

Upstream code and original data: [https://github.com/Roche/OligoGym](https://github.com/Roche/OligoGym)