---
license: mit
tags:
  - peptide
  - blood-brain-barrier
  - bbb
  - classification
  - esm
library_name: bbb-models
pipeline_tag: text-classification

base_model: manumartinm/bbb-classifier
---

# BBB Classifier — `exp06_esm_tab_mlp_aug`

Sequence/tabular BBB permeability classifier for peptide candidates. Fuses ESM-2 embeddings with physicochemical descriptors and isotonic-calibrated probabilities.

## Model summary

| Field | Value |
|-------|-------|
| Experiment | `exp06_esm_tab_mlp_aug` |
| Architecture | `esm_tab_mlp` |
| ESM dim | `128` |
| Hidden dim | `256` |
| PR-AUC (calibrated) | `0.906775452423001` |
| ROC-AUC (calibrated) | `0.9185328185328185` |
| MCC (calibrated) | `0.6435046074523986` |

## Files

- `checkpoints/best.ckpt` — PyTorch weights
- `calibrators/calibrator.pkl` — isotonic calibration
- `train_metadata.json` — experiment + data config for inference
- `metrics.json` — validation metrics

## Usage

Install the project and download this repo:

```bash
git clone https://github.com/your-org/TFG.git
cd TFG/bbb_models && uv sync
hf download manumartinm/bbb-classifier --local-dir ./bbb-classifier
```

Score candidates (input must include sequence + tabular descriptor columns used at training time):

```bash
uv run python scripts/classifier/predict.py \
  --run-dir ./bbb-classifier \
  --input candidates.parquet \
  --output scored.parquet
```

## Dataset

Trained on the TFG BBB peptide dataset. Companion dataset: [`manumartinm/bbb-peptides`](https://huggingface.co/datasets/manumartinm/bbb-peptides).

## Citation

If you use this model, cite the TFG BBB peptide modeling pipeline.