# Cryptic-Binding

Cryptic protein binding sites from CryptoBank + apo-holo pair mining.
Filtered for CDR3 compatibility and NMA open-state stability, then
clustered by Zernike pocket-shape descriptors.

## Files

- `cryptic_sites_final.csv` (1990 rows) - all surviving sites
- `cryptic_sites_centroids.csv` (1 rows) - one per cluster
- `farm_report.md` - filter funnel + top sites

**Evidence:** experimental=1990, computational=0

## Columns

`site_id`, `pdb_id`, `uniprot_id`, `chain`, `source`, `evidence_tier`, `pocket_volume`, `asa_a2`, `depth_a`, `width_a`, `polar_fraction`, `compatibility_score`, `nma_pocket_variance`, `open_state_stable`, `cluster_id`, `is_centroid`, `notes`

## Usage

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

path = hf_hub_download('Nanoflow/nanoflow-data', 'Cryptic-Binding/cryptic_sites_centroids.csv',
                       repo_type='dataset')
df = pd.read_csv(path)
```

Structures are referenced by PDB ID; download from RCSB as needed.