strand-ai/variantformer-1000g

---
license: cc-by-4.0
task_categories:
- tabular-classification
tags:
- genomics
- variants
- 1000-genomes
- gene-expression
- variantformer
size_categories:
- 100G<n<1T
---

# VariantFormer 1000 Genomes Dataset

Gene expression predictions from VariantFormer for 538 samples from the 1000 Genomes Project.

## Dataset Structure

```
├── manifest.csv # Sample metadata (population, sex)
├── predictions/ # VariantFormer predictions
│ └── {sample_id}.parquet
└── vcf/ # Per-sample VCF files
├── {sample_id}.vcf.gz
└── {sample_id}.vcf.gz.tbi
```

## Files

- **manifest.csv**: Sample metadata with columns: `sample_id`, `population`, `superpopulation`, `sex`
- **Parquet files**: VariantFormer gene expression predictions (~446 MB per sample, ~240 GB total)
- **VCF files**: Variant calls per sample with tabix indexes (~380 GB total)

## Usage

```python
import pandas as pd
from huggingface_hub import hf_hub_download, snapshot_download

# Download and load sample manifest
manifest_path = hf_hub_download(
repo_id="strand-ai/variantformer-1000g",
filename="manifest.csv",
repo_type="dataset"
)
manifest = pd.read_csv(manifest_path)

# Download predictions for a single sample
pred_path = hf_hub_download(
repo_id="strand-ai/variantformer-1000g",
filename="predictions/HG00418.parquet",
repo_type="dataset"
)
df = pd.read_parquet(pred_path)

# Download ALL data locally (~620 GB)
snapshot_download(
repo_id="strand-ai/variantformer-1000g",
repo_type="dataset",
local_dir="./variantformer-1000g"
)
```

## Interactive Explorer

Explore the data interactively at [strandai.bio/1000g-variantformer](https://strandai.bio/1000g-variantformer)

## Citation

If you use this dataset, please cite:

```
@dataset{strand_variantformer_1000g,
title={VariantFormer 1000 Genomes Predictions},
author={Strand AI},
year={2026},
url={https://huggingface.co/datasets/strand-ai/variantformer-1000g}
}
```

## License

This dataset is released under CC-BY-4.0 for research use.

## Contact

Questions? Email us at [[e-mail]](mailto:[e-mail])