---
license: mit
---

# ProtCompass Embeddings

Pre-computed protein embeddings from 70+ encoders across 13 downstream tasks.

## Dataset Structure

```
embeddings/
├── secondary_structure/     # CB513 dataset (29 GB)
├── mutation_effect/         # ProteinGym DMS assays (4.5 GB)
├── contact_prediction/      # ProteinNet (2.9 GB)
├── stability/              # TAPE stability (1.6 GB)
├── ppi_site/               # PPI site prediction (1.4 GB)
├── fluorescence/           # GFP fluorescence (841 MB)
├── metal_binding/          # Metal binding sites (570 MB)
├── go_bp/                  # GO Biological Process (214 MB)
├── go_mf/                  # GO Molecular Function (68 MB)
├── remote_homology/        # SCOPe fold classification (20 MB)
├── ec_classification/      # Enzyme classification (18 MB)
├── membrane_soluble/       # Membrane/soluble (17 MB)
└── subcellular_localization/ # Subcellular location (17 MB)
```

## File Format

Each encoder directory contains:
- `train_embeddings.npy`: Training set embeddings (N × D)
- `test_embeddings.npy`: Test set embeddings (M × D)
- `train_labels.npy`: Training labels
- `test_labels.npy`: Test labels
- `train_ids.txt`: Protein IDs for training set
- `test_ids.txt`: Protein IDs for test set
- `meta.json`: Metadata (encoder name, dimensions, dataset info)

## Usage

```python
import numpy as np
from huggingface_hub import hf_hub_download

# Download specific encoder embeddings
train_emb = np.load(hf_hub_download(
    repo_id="Anonymoususer2223/ProtCompass_Embeddings",
    filename="embeddings/mutation_effect/esm2/train_embeddings.npy",
    repo_type="dataset"
))

test_emb = np.load(hf_hub_download(
    repo_id="Anonymoususer2223/ProtCompass_Embeddings",
    filename="embeddings/mutation_effect/esm2/test_embeddings.npy",
    repo_type="dataset"
))

# Use for downstream tasks
from sklearn.linear_model import Ridge
model = Ridge()
model.fit(train_emb, train_labels)
score = model.score(test_emb, test_labels)
```

## Encoders Included

### Sequence Encoders (8)
ESM2 (650M, 150M, 35M), ESM1b, ESM3, ProtTrans, ProST-T5, ProteinBERT-BFD, Ankh

### Structure Encoders (50+)
GearNet, GCPNet, EGNN, GVP, IPA, TFN, SchNet, DimeNet, MACE, CDConv, ProteinMPNN, PottsMP, dMaSIF

### Multimodal Encoders (5)
SaProt, ESM-IF, FoldVision

### Baselines
Random, Length, Torsion, One-hot, BLOSUM

## Dataset Statistics

- **Total size**: 41 GB
- **Total encoders**: 70+
- **Total tasks**: 13
- **Total proteins**: ~500K across all tasks

## Citation

If you use these embeddings, please cite:

```bibtex
@article{protcompass2026,
  title={ProtCompass: Systematic Evaluation of Protein Structure Encoders},
  author={Your Name et al.},
  journal={NeurIPS},
  year={2026}
}
```

## License

MIT License

## Contact

For questions or issues, please open an issue on the repository.