---
title: TEM1 Target Finder
emoji: 🧪
colorFrom: blue
colorTo: green
sdk: gradio
sdk_version: 4.40.0
app_file: app.py
pinned: false
---


# TEM-1 Target Finder (Gradio on Hugging Face Spaces)

This Space predicts small-molecule binding strength (pAff = −log10 Kd) to **TEM-1 β-lactamase** using:
- **Protein** embeddings from ESM-2 (35M)
- **Ligand** embeddings from ChemBERTa
- Lightweight **XGBoost** regressor for pAff and **Logistic Regression** for binder probability, with isotonic calibration

## How to deploy

1. Create a new **Gradio** Space.
2. Upload the following files:
   - `app.py`
   - `requirements.txt`
   - *(optional)* `tem1_clean.csv` — your dataset with at least two columns:
     - `smiles` (string)
     - `pAff` (float; e.g., 6.0 for 1 µM)

3. If `tem1_clean.csv` is present, the Space will **train heads on startup** and show real metrics.  
   If not present, it runs in **demo mode** with heuristic heads (usable but not scientifically calibrated).

> You can place the CSV at `./tem1_clean.csv` or `./data/tem1_clean.csv`.

## Notes

- Runs on **CPU** by default. First startup downloads ESM-2 and ChemBERTa from Hugging Face.
- Batch plots support up to 100 SMILES pasted as lines or comma/semicolon separated.
- pAff threshold **≥6** is used as a binder proxy (≈ ≤1 µM).