---
license: mit
tags:
- biology
- CRISPR
---




# CRISPR Efficiency Predictor

A deep learning model for predicting CRISPR-Cas9 editing efficiency based on DNA sequences and epigenetic features. This model integrates sequence data and epigenetic signals to provide highly accurate predictions of CRISPR editing efficiency.

---

## Model Details

- **Model Type**: Convolutional Neural Network (CNN)
- **Input Features**:
  - **DNA Sequence**: 23-base target sequence, one-hot encoded.
  - **Epigenetic Features**:
    - CTCF (Transcription factor binding sites)
    - DNase (Chromatin accessibility)
    - H3K4me3 (Histone modification marker)
    - RRBS (Methylation marker)
- **Output**: A single efficiency score indicating the likelihood of successful CRISPR editing for the given input.

---

## Training and Evaluation

### **Training Details**
- **Dataset**: [DeepCRISPR](https://github.com/bm2-lab/DeepCRISPR)
  - Citation: Guohui Chuai, Qi Liu et al. *DeepCRISPR: optimized CRISPR guide RNA design by deep learning*. 2018 (Manuscript submitted).
- **Framework**: TensorFlow/Keras
- **Optimizer**: Adam
- **Loss Function**: Mean Squared Error (MSE)

### **Evaluation Metrics**
| Metric                       | Value    |
|------------------------------|----------|
| R-squared (R²)               | 0.9754   |
| Pearson Correlation Coefficient | 0.9876   |
| Mean Residual                | -0.0003  |
| Residual Standard Deviation  | 0.0032   |

---