EHNN: Enhanced Hybrid Neural Network for CRISPR–Cas9 gRNA Off-Target Prediction

# EHNN: Enhanced Hybrid Neural Network for CRISPR–Cas9 Off-Target Prediction ## 📌 Description EHNN is a hybrid deep-learning pipeline that combines XOR-based sequence encoding, k-mer context, PAM classification, and 51 engineered biochemical features to predict CRISPR–Cas9 off-target cleavage events. It is the **first architecture** that jointly integrates: * **XOR mismatch maps*** **k-mer contexts*** **PAM categories*** **51 engineered biochemical sequence features** The model fuses **CNN–LSTM–MLP** modules and achieves robust accuracy and calibration across multiple public CRISPR datasets. --- ## 🔗 How to cite If you use this code or data, please cite: > Sayed A., Ullah K., Khan M., Alakhtar R., Alsobhi H. > *EHNN: Enhanced Hybrid Neural Network for CRISPR–Cas9 gRNA Off-Target Prediction*. > Zenodo, https://doi.org/10.5281/zenodo.18165347 --- 📂 Project Structure original_dataset/ # Raw input CSVs (on_seq, off_seq, label) feature_dataset/ # Extracted biochemical and sequence features normalize_dataset/ # Min-Max scaled features encoded_dataset/ # XOR / k-mer / PAM encoded tensors Final_Results/ # Consolidated outputs & manuscript figures ├── full_results_EHNN.csv ├── figures/ # ROC, PR, calibration plots └── tables/ # CSV / TSV for paper tables requirements.txt LICENSE README.md ehnn_pipeline.py # Main entry point (pre-processing → training → eval) **Files in `original_dataset/`** - `CIRCLE_seq.csv`- `Doench.csv`- `GUIDE-Seq_Kleinstiver.csv`- `GUIDE-Seq_Listgarten.csv`- `Hek293t.csv`- `Hek293t_K562.csv`- `K562.csv`- `SITE-Seq.csv` 📄 LicenseMIT — see LICENSE.