---
license: apache-2.0
tags:
- biology
- genomics
- dna
- eccdna
size_categories:
- 10K<n<1M
task_categories:
- token-classification
---

# Real vs. Pseudo-eccDNA Discrimination (Homo sapiens)

This dataset supports the **Real vs. Pseudo-eccDNA Discrimination** task for human eccDNA.  
The goal is to train models that can distinguish true eccDNA sequences from pseudo-eccDNAs  
randomly extracted from linear genomic regions with matched length distributions.

Each entry contains:
- `sequence`: raw eccDNA sequence (A/T/C/G)
- `label`:  
  - `1` → Real eccDNA  
  - `0` → Pseudo-eccDNA (negative control)

---

## 📁 Folder Structure
<pre>
real_vs_pseudo_eccdna_discrimination_human/
├── data/
│   └── real_vs_pseudo_eccdna_discrimination_human.csv
└── README.md
</pre>

---

## Task Description

True eccDNAs are experimentally verified circular DNA molecules,  
whereas pseudo-eccDNAs are generated by randomly extracting linear genomic segments  
to match the true eccDNA length distribution.  
This task assesses a model’s ability to capture **circular topology** and **regulatory context**  
beyond simple sequence composition.

---

## Citation
If you use this dataset, please cite:
<pre><code class="language-python">

@inproceedings{liu2025eccdnamamba,
  title={eccDNAMamba: A Pre-Trained Model for Ultra-Long eccDNA Sequence Analysis},
  author={Zhenke Liu and Jien Li and Ziqi Zhang},
  booktitle={ICML 2025 GenBio Workshop},
  year={2025},
  url={https://openreview.net/forum?id=56xKN7KJjy}
}
</code></pre>