README.md

---
license: cc-by-nc-sa-4.0
tags:
- genomics
- dna
- benchmark
pretty_name: GLMap Probe Panels
---

# GLMap Probe Panels

Prebuilt DNA probe panels for [GLMap](https://github.com/ai4nucleome/GLMap)
— a training-free framework for profiling genomic language models by their
likelihood responses over a fixed panel of DNA sequences.

## Files

| File | Description |
|---|---|
| `main_panel.parquet` | 10,000-probe biological panel (14 functional elements × 4 categories) |
| `panel_manifest.json` | Per-element / per-dataset counts and sampling seeds |
| `panel_summary.md` | Human-readable panel composition summary |

## Usage

```python
import glmap
panel = glmap.load_panel()   # auto-downloads from this dataset if not local
```

Or directly:

```python
import pandas as pd
from huggingface_hub import hf_hub_download
path = hf_hub_download("Tim419/GLMap-panels", "main_panel.parquet", repo_type="dataset")
panel = pd.read_parquet(path)
```

## License

**CC-BY-NC-SA-4.0.** The panel was assembled from three upstream benchmarks:

- GUE (DNABERT-2) — Apache-2.0
- Plant Genomic Benchmark (ArgoNT) — CC-BY-NC-SA-4.0
- DNA Foundation Benchmark (DFB) — Apache-2.0

Because the panel contains 1,600 probes drawn from PGB, the combined
artefact inherits PGB's CC-BY-NC-SA-4.0 license (ShareAlike). For
non-commercial research use with attribution. Please cite all three
upstream benchmarks plus the GLMap paper.