README.md

---
library_name: tedbench
tags:
  - protein
  - structure-sequence
  - fold-classification
  - tedbench
  - saprot
pipeline_tag: other
license: bsd-3-clause
---

# TEDBench — SaProt-650M fine-tuned on TEDBench

Backbone: SaProt-650M (33 layers, hidden dim 1280). Requires [Foldseek](https://github.com/steineggerlab/foldseek) for structure-aware tokens.

Fine-tuned on [TEDBench](https://github.com/BorgwardtLab/TEDBench) for protein
fold classification into 965 CATH topology (T-level) classes (ICML 2026).

## Usage

```python
import sys
sys.path.insert(0, "baselines")  # from repo root

from pathlib import Path
import torch
from models.saprot_classifier import SaProtClassifier
from omegaconf import OmegaConf
from huggingface_hub import snapshot_download

local_dir = Path(snapshot_download("TEDBench/saprot-650M-ft"))
with open(local_dir / "config.json") as f:
    import json
    cfg = OmegaConf.create(json.load(f))

model = SaProtClassifier(cfg)
sd = torch.load(local_dir / "pytorch_model.bin", map_location="cpu", weights_only=False)
model.load_state_dict(sd)
model.eval()
```

Or pass the repo ID directly to the test script:

```bash
python baselines/saprot_test_ted.py train.ckpt_path=TEDBench/saprot-650M-ft
```

## Citation

```bibtex
@inproceedings{chen2026tedbench,
  title={Protein Fold Classification at Scale: Benchmarking and Pretraining},
  author={Chen, Dexiong and Manolache, Andrei and Niepert, Mathias and Borgwardt, Karsten},
  booktitle={Proceedings of the 43rd International Conference on Machine Learning},
  year={2026}
}
```