README.md

---
license: apache-2.0
task_categories:
  - text-classification
  - token-classification
  - text-generation
language:
  - en
tags:
  - genomics
  - bioinformatics
  - kluyveromyces-marxianus
  - biobert
  - functional-genomics
  - pangenome
  - yeast
  - biomistral
size_categories:
  - 10K<n<100K
---

# 🧬 Kluyveromyces marxianus - BioBERT-Optimized Genomics Dataset

<div align="center">
  <img src="https://img.shields.io/badge/BioBERT-Optimized-green" alt="BioBERT"/>
  <img src="https://img.shields.io/badge/Chunks-2-brightgreen" alt="Chunks"/>
  <img src="https://img.shields.io/badge/Quality-0.855-yellow" alt="Quality"/>
  <img src="https://img.shields.io/badge/512_tokens-Ready-blue" alt="512 tokens"/>
</div>

## 📋 Overview

**2 semantically-optimized chunks** for BioBERT fine-tuning, processed with **BiOMistral-7B** quantum chunking engine.

### Key Features

✅ **Deep Semantic Coherence** (avg: 1.000)
✅ **Optimal Token Size** (~512 tokens per chunk)
✅ **Quality Filtering** (2 high-quality chunks)
✅ **Entity Extraction** (genes, proteins, pathways, conditions)
✅ **BioBERT-Ready** (1 chunks ≤512 tokens)

## 🎯 Research Context

**PhD Thesis**: *Functional Genomics of Robust Linear Yeasts (Kluyveromyces marxianus) using BioBERT and Pangenome Methodology for Identifying Key Survival Genes in Severe Gut Conditions*

### Applications

- 🧬 Gene survival mechanism discovery
- 🔬 Metabolic pathway analysis
- 🦠 Stress response characterization
- 📊 Pangenome comparative genomics
- 🤖 AI-driven gene function prediction

## 📊 Dataset Statistics

| Metric | Value |
|--------|-------|
| **Total Chunks** | 2 |
| **High-Quality (≥0.7)** | 2 (100.0%) |
| **BioBERT-Ready** | 1 (50.0%) |
| **Avg Coherence** | 1.0000 |
| **Avg Density** | 0.7108 |
| **Avg Quality** | 0.8554 |
| **Processing Date** | 2025-11-01T08:29:54.885911 |

## 🧬 Data Structure

```json
{
  "chunk_id": "train_0_0",
  "global_id": 0,
  "text": "Optimized text content...",
  "token_count": 487,
  "semantic_coherence": 0.8234,
  "information_density": 0.7156,
  "quality_score": 0.7695,
  "entities": {
    "genes": ["ABC1", "XYZ2"],
    "proteins": ["hexokinase"],
    "pathways": ["glycolysis"],
    "organisms": ["marxianus"],
    "conditions": ["acid", "gut"]
  },
  "biobert_ready": true
}
```

## 📁 Files

- `chunks_complete.json` - All chunks with metadata (2 chunks)
- `chunks_high_quality.json` - Premium subset (2 chunks)
- `chunks.jsonl` - Streaming format
- `statistics.json` - Processing statistics
- `analysis_dashboard.png` - Visual analytics
- `interactive_dashboard.html` - Interactive Plotly dashboard
- `entities_wordcloud.png` - Entity visualization

## 🚀 Quick Start

### Load Dataset

```python
import json

# High-quality subset
with open('chunks_high_quality.json', 'r') as f:
    chunks = json.load(f)

print(f"Loaded {len(chunks):,} high-quality chunks")
```

### BioBERT Fine-tuning

```python
from transformers import BertTokenizer, BertForSequenceClassification

model = BertForSequenceClassification.from_pretrained(
    "dmis-lab/biobert-large-cased-v1.1",
    num_labels=YOUR_CLASSES
)

tokenizer = BertTokenizer.from_pretrained(
    "dmis-lab/biobert-large-cased-v1.1"
)

# Chunks are already 512-token optimized!
for chunk in chunks:
    encoding = tokenizer(
        chunk['text'],
        truncation=True,
        max_length=512,
        padding='max_length',
        return_tensors='pt'
    )
    # Train your model...
```

### Filter by Quality

```python
# Get excellent chunks (≥0.8)
excellent = [c for c in chunks if c['quality_score'] >= 0.8]

# Get gut-stress related chunks
gut_chunks = [
    c for c in chunks
    if any(cond in c['entities']['conditions']
           for cond in ['gut', 'acid', 'bile'])
]
```

## 📈 Quality Metrics

**Semantic Coherence** (0-1): Thematic consistency using BiOMistral embeddings
**Information Density** (0-1): Vocabulary diversity (unique/total words)
**Quality Score** (0-1): Combined metric `(coherence + density) / 2`

### Quality Tiers

- **Excellent** (≥0.8): Highly focused, rich content
- **Good** (0.7-0.8): Strong coherence and diversity
- **Acceptable** (0.5-0.7): Moderate quality
- **Low** (<0.5): Basic content

## 🔬 Processing Pipeline

1. **BiOMistral-7B Loading** - 4-bit/FP16 quantization
2. **Semantic Analysis** - Deep understanding via embeddings
3. **Intelligent Chunking** - Boundary detection at natural breaks
4. **Entity Extraction** - Automatic identification
5. **Quality Scoring** - Multi-metric assessment
6. **Visualization** - Interactive and static analytics

## 🎓 Citation

```bibtex
@dataset{kmx_chunks_2024,
  title = {Kluyveromyces marxianus BioBERT-Optimized Chunks},
  author = {Quantum Chunking System},
  year = {2024},
  publisher = {HuggingFace},
  url = {https://huggingface.co/datasets/Milad96/Kluyveromyces-marxianus-chunks}
}
```

## 📜 License

Apache 2.0 - Free for commercial and research use

## 🤝 Source

**Original Dataset**: [Milad96/Kluyveromyces-marxianus](https://huggingface.co/datasets/Milad96/Kluyveromyces-marxianus)

**Processing**: BiOMistral-7B Quantum Chunking System

## 🔗 Resources

- [BioBERT](https://huggingface.co/dmis-lab/biobert-large-cased-v1.1)
- [BiOMistral-7B](https://huggingface.co/BioMistral/BioMistral-7B)
- [Transformers](https://github.com/huggingface/transformers)

---

**Generated**: 2025-11-01 08:30:12
**System**: BiOMistral Quantum Chunking v1.0
**Target**: BioBERT-Large Fine-tuning