---
language:
- en
tags:
- protein
- bioinformatics
- protein sequence
- protein function
- enzyme
- uniprot
- sequence-to-text
- protein annotation
- biological process
- molecular function
- cellular component
configs:
- config_name: default
  description: Protein sequences with functional descriptions and annotations.
  data_files:
  - split: train
    path: NL2Protein.csv
  features:
  - name: ID
    dtype: string
  - name: Text
    dtype: string
  - name: Protein
    dtype: string
  - name: Species
    dtype: string
  - name: Family
    dtype: string
  viewer:
    type: table
    rows_per_page: 10
size_categories:
- 100K<n<1M
---
# NL2Protein Dataset
## I. Dataset Introduction
This dataset provides aligned pairs of natural language descriptions and protein
sequences, where the descriptions integrate protein family information, length
constraints, and Gene Ontology–based functional relations. The dataset is
intended for studying natural language–guided protein sequence generation and
the semantic alignment between textual protein descriptions and sequence space.

The data in this dataset are derived from the UniProt database. Protein
sequences, organism information, protein family annotations, and associated GO
terms were downloaded from UniProt and subsequently integrated, filtered, and
reformatted into a unified text–sequence paired dataset for research use.

Each entry includes:
- A protein identifier
- A structured functional description
- A full-length amino acid sequence
- The source organism (species)
- A protein family annotation

The functional description contains:
- Protein family
- Protein length range and size category
- Biological processes
- Molecular functions
- Cellular components
- Parent–child (is_a) relationships between GO terms
- Cross-ontology relationships, such as:
  - part_of
  - occurs_in
  - regulates / positively_regulates / negatively_regulates

This dataset is designed for natural language–to–protein sequence generation.
Given a natural language description that specifies functional, and
biological constraints—such as protein family, length range, Gene Ontology (GO)
terms from the Biological Process, Molecular Function, and Cellular Component
namespaces, as well as their parent–child and cross-ontology relationships—the
task is to directly generate a full-length amino acid sequence that is
consistent with the provided description. The dataset enables training and
evaluation of protein language models on text-conditioned sequence generation,
supporting research in controllable protein design and semantic alignment
between natural language descriptions and protein sequence space.

## Ⅱ. Data Structure
Each sample consists of the following fields:
- ID: Protein identifier (UniProt-style accession)
- protein: Amino acid sequence
- Text: Functional description text
- species: Source organism of the protein (UniProt Organism field)
- family: Protein family annotation derived from UniProt

| ID | Text | Protein | Species | Family |
|----|---------|---------|--------|------|
| A0A017SPL2 | Belongs to the Tryptophan dimethylallyltransferase family.<br>The protein length falls within the range of 301–500 amino acids, belonging to the category of Large proteins.<br>It participates in biological processes such as alkaloid metabolic process.<br>It exhibits molecular functions including prenyltransferase activity. | MQPYHTLSRVLPFPDANQKAWWDKLGPMLLKAMQS<br>QGYDTEAQYAQLGMVYKCVLPYLGEFPTVENDATRWK<br>SFLCPYGIPIEPSLNISQGILRYAFEPIGPDVGTEKDPQN<br>MN | Accumulibacter regalis | Tryptophan dimethylallyltransferase family |
## Ⅲ. Dataset Loading
```python
from datasets import load_dataset
dataset = load_dataset(
    "csv",
    data_files="NL2Protein.csv"
)
print(dataset)
print(dataset["train"][0])
```

## BibTeX