|
|
--- |
|
|
license: apache-2.0 |
|
|
task_categories: |
|
|
- question-answering |
|
|
- text-generation |
|
|
language: |
|
|
- zh |
|
|
- en |
|
|
tags: |
|
|
- microbiology |
|
|
- medical |
|
|
- qa |
|
|
- infectious-diseases |
|
|
- pathogens |
|
|
size_categories: |
|
|
- 100K<n<1M |
|
|
--- |
|
|
|
|
|
# π¦ Microbiology QA Complete Dataset v2.1 |
|
|
|
|
|
The most comprehensive microbiology question-answering dataset, combining LLM-generated content, real clinical reports, and filtered open-source medical data. |
|
|
|
|
|
## π Dataset Statistics |
|
|
|
|
|
| Split | Samples | |
|
|
|-------|---------| |
|
|
| Train | 278,073 | |
|
|
| Validation | 15,448 | |
|
|
| Test | 15,450 | |
|
|
| **Total** | **308,971** | |
|
|
|
|
|
## π Data Sources |
|
|
|
|
|
### 1. LLM Textbook QA (15,487 samples) |
|
|
High-quality QA pairs extracted by Qwen-Plus from authoritative textbooks: |
|
|
- Sanford Guide to Antimicrobial Therapy: 3,350 |
|
|
- Manual of Clinical Microbiology (Vol.1 & 2): 7,665 |
|
|
- Harrison's Principles of Infectious Diseases: 4,472 |
|
|
|
|
|
### 2. CSV Report Analysis QA (22,842 samples) β¨ Enhanced |
|
|
From 9,990 real microbiology metagenomic reports: |
|
|
- Species identification & abundance analysis |
|
|
- Pathogen detection & clinical significance |
|
|
- Microbiome diversity assessment |
|
|
- **Enhancement**: Deduplicated 1,183 entries + Question diversification (5 variants per high-frequency question via Qwen-Plus) |
|
|
|
|
|
### 3. Filtered Open-source Medical Data (270,642 samples) |
|
|
Filtered from 546K+ open-source medical QA using 272 microbiology keywords: |
|
|
|
|
|
| Source | Samples | Language | Content | |
|
|
|--------|---------|----------|---------| |
|
|
| PubMedQA | 187,403 | English | Biomedical research QA | |
|
|
| MedMCQA | 49,226 | English | Medical multiple choice | |
|
|
| CMtMedQA | 27,377 | Chinese | Doctor-patient dialogues | |
|
|
| MedQA | 6,637 | English | USMLE-style questions | |
|
|
|
|
|
## π Data Format |
|
|
|
|
|
```json |
|
|
{ |
|
|
"instruction": "Is reflective functioning associated with clinical symptoms?", |
|
|
"input": "Mentalization is the capacity to understand behavior...", |
|
|
"output": "These findings indicate that RF is associated with...", |
|
|
"source": "pubmedqa/train_pqa_unlabeled.jsonl", |
|
|
"category": "pubmed_qa", |
|
|
"matched_keywords": ["uti"], |
|
|
"data_source": "opensource" |
|
|
} |
|
|
``` |
|
|
|
|
|
## π― Topic Coverage |
|
|
|
|
|
- **Pathogens**: Bacteria, viruses, fungi, parasites |
|
|
- **Infections**: Pneumonia, sepsis, UTI, etc. |
|
|
- **Antimicrobials**: Antibiotics, antivirals, antifungals |
|
|
- **Resistance**: MRSA, VRE, CRE, ESBL |
|
|
- **Diagnostics**: Culture, PCR, mNGS, mass spectrometry |
|
|
- **Microbiome**: Gut flora, dysbiosis |
|
|
|
|
|
## π¦ Usage |
|
|
|
|
|
```python |
|
|
from datasets import load_dataset |
|
|
|
|
|
dataset = load_dataset("xingqiang/microbiology-qa-complete") |
|
|
print(f"Training samples: {len(dataset['train']):,}") |
|
|
# Output: Training samples: 278,073 |
|
|
|
|
|
# Filter by data source |
|
|
llm_data = dataset['train'].filter(lambda x: x['data_source'] == 'llm_textbook') |
|
|
opensource_data = dataset['train'].filter(lambda x: x['data_source'] == 'opensource') |
|
|
``` |
|
|
|
|
|
## π v2.1 Updates |
|
|
|
|
|
- CSV data deduplication: Removed 1,183 duplicate QA pairs |
|
|
- Question diversification: Generated 5 expression variants for high-frequency questions |
|
|
- Reduced question reuse rate for better training diversity |
|
|
|
|
|
## π Related Datasets |
|
|
|
|
|
- [microbiology-llm-qa](https://huggingface.co/datasets/xingqiang/microbiology-llm-qa) - Pure LLM-generated (39K) |
|
|
- [microbiology-qa-v1](https://huggingface.co/datasets/xingqiang/microbiology-qa-v1) - Earlier version (226K) |
|
|
|
|
|
## π
Version Info |
|
|
|
|
|
- **Version**: 2.1 (Enhanced) |
|
|
- **Date**: December 2024 |
|
|
- **Project**: [DeepMicroPath](https://github.com/chenxingqiang/DeepMicroPath) |
|
|
|
|
|
## π License |
|
|
|
|
|
Apache 2.0 |
|
|
|