---
dataset_name: IDP-Euka-90
pretty_name: IDP-Euka-90
---

# IDP-Euka-90

**IDP-Euka-90** is a collection of eukaryotic protein sequences curated for representation learning and downstream analysis of **intrinsically disordered proteins/regions (IDPs/IDRs)**.
These sequences were obtained by running [Metapredict V3](https://pubmed.ncbi.nlm.nih.gov/34480923/) on all the available eukaryota proteomes from [UniProt](https://www.uniprot.org), which extracted all the IDP regions in each proteome. The sequences we subsequently clustered at 90% with [mmseqs2](https://github.com/soedinglab/MMseqs2) to remove all near duplicates.
The HF dataset is distributed as a train/val split and the backup csv has a **`split` column** prepared.

---

## Contents

- **Columns**
  - `sequence` — protein amino-acid sequence (single-letter codes)

- **Format**
  - Hosted on the Hugging Face Hub as an Arrow/CSV-backed dataset.

---

## Quick start

```python
from datasets import load_dataset

repo_id = "InstaDeepAI/IDP-Euka-90"
ds = load_dataset(repo_id)
print(ds)
print(ds.features)