---

license: mit
language:
  - en
tags:
  - biomedical
  - clinvar
  - genomic-variants
  - instruction-tuning
  - text-generation
task_categories:
  - text-generation
  - tabular-classification
size_categories:
  - 10K<n<100K
configs:
  - config_name: tabular
    data_files:
      - split: train
        path: train.csv
      - split: validation
        path: valid.csv
      - split: test
        path: test.csv
  - config_name: instruction
    data_files:
      - split: train
        path: biogpt_train.csv
      - split: validation
        path: biogpt_valid.csv
      - split: test
        path: biogpt_test.csv

---

**Year:** 2025  
**License:** MIT  
**Author:** Sepideh Moafi 

---

# PathogenAgentAI Instruction Dataset

## Dataset Description

A ClinVar-derived dataset developed as part of the **PathogenAgentAI** research software project. The dataset is released in two parallel formats:

- **Tabular version** (`train.csv`, `valid.csv`, `test.csv`) — structured genomic-variant data for classical ML and analysis.
- **BioGPT instruction version** (`biogpt_train.csv`, `biogpt_valid.csv`, `biogpt_test.csv`) — instruction-style data for fine-tuning biomedical language models.

## Dataset Structure

### Splits

| Split | Tabular file | Instruction file |
|---|---|---|
| Train | `train.csv` | `biogpt_train.csv` |
| Validation | `valid.csv` | `biogpt_valid.csv` |
| Test | `test.csv` | `biogpt_test.csv` |

### Tabular Fields

- `GeneSymbol` (string): Gene symbol (e.g., AP5Z1, ZNF592)
- `ClinicalSignificance` (string): Clinical significance (e.g., Pathogenic, Likely pathogenic, Uncertain significance)
- `Assembly` (string): Genome assembly (e.g., GRCh38)
- `Chromosome` (string): Chromosome number
- `Start` (int64): Variant start position

### Instruction Fields

The `biogpt_*.csv` files are formatted for instruction tuning of biomedical language models (e.g., BioGPT), derived from the same ClinVar source data.

## Source Data

Derived from **ClinVar**, a publicly available archive of reports of the relationships among human variations and phenotypes.

## Intended Use

- Fine-tuning biomedical language models (BioGPT and similar)
- Research on genomic variant interpretation
- Classical ML on structured variant data
- Reproducible biomedical AI experiments

**Not** intended for clinical diagnosis, treatment decisions, or direct clinical use.

## Licensing

Released under the **MIT License**.

## Citation

If you use this dataset in research, please cite:

> BioGPT-ClinVar: Parameter-Efficient Fine-Tuning of a Biomedical Foundation Model.  
> Research Square, 2026. DOI: 10.21203/rs.3.rs-10196893/v1

## Contact

**Sepideh Moafi**  
GitHub: [AIResearcher20](https://github.com/AIResearcher20)  
Hugging Face: [Sepideh2027](https://huggingface.co/Sepideh2027)