---
dataset_info:
  features:
  - name: protein_name
    dtype: string
  - name: species
    dtype: string
  - name: sequence
    dtype: string
  - name: annotation
    dtype: string
  - name: messages
    list:
    - name: content
      dtype: string
    - name: role
      dtype: string
  splits:
  - name: train
    num_bytes: 63705781
    num_examples: 21332
  - name: test
    num_bytes: 11945580
    num_examples: 4000
  download_size: 45260837
  dataset_size: 75651361
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: test
    path: data/test-*
---

## Dataset Transformation Summary

**Original Dataset**: `monsoon-nlp/primate-proteins`  
**Transformed Dataset**: `pkanithi/primate-proteins`

## Changes Made

### Added `messages` Column
- Added a new `messages` column in ChatML format
- Each example now contains a conversation structure with system, user, and assistant messages

### Data Filtering
- Filtered out proteins with no annotation (`annotation != None`)
- Ensures all examples have valid ground truth annotations

### Dataset Splitting
- Split the original 'train' split into train and test sets
- Test set: 4,000 examples
- Train set: Remaining examples
- Used seed=42 for reproducible splits

### Chat Format Structure
Each example now has a `messages` array with:
1. **System message**: Instructions for protein annotation assistance
2. **User message**: Protein entry with name, species, and sequence
3. **Assistant message**: Original annotation from the dataset

The user message format:
```
Here is the protein entry:
- protein_name: [protein_name]
- species: [species]  
- sequence: [sequence]
```

## Result
The dataset is now in ChatML format suitable for supervised fine-tuning with train/test splits, while preserving all original protein annotation content.