---
license: mit
language: [en]
tags:
- seq2seq
- dna
- protein
- eb502
---

# DNA → Amino Acid Translation Dataset (EB502 HW1)

## Description

This dataset contains randomly generated DNA sequences (A, C, G, T)
of fixed length (e.g., 30 nucleotides), and their corresponding
amino acid sequences translated using the standard genetic code.

Each example is a simple **sequence-to-sequence** pair:

- `src`: input DNA sequence (string)
- `tgt`: output amino acid sequence (string)

Stop codons (*TAA, TAG, TGA*) terminate translation,
and empty translations are filtered out.

This dataset is created as part of **KAIST EB502 (Seq2Seq Data HW1)**.

## Dataset Structure

- **Fields**
  - `src`: DNA sequence, composed of 'A', 'C', 'G', 'T'
  - `tgt`: translated amino acid sequence, using standard codon table

- **Splits**
  - `train` (~80% of samples)
  - `validation` (~10% of samples)
  - `test` (~10% of samples)

## Intended Use

This dataset is intended for educational purposes, to practice:

- Sequence-to-sequence data preparation
- Training simple translation models
- Using Hugging Face Datasets and Hub

## License

MIT License.