CRISPROff: sgRNA Off-Target Prediction with Boosting Deep Learning

This is the official repository of the paper: CRISPROff for sgRNA off-target prediction based on boosting deep learning.

PREREQUISITE:

CRISPROFF was conducted using TensorFlow version 2.3.2 and Python version 3.6.

To set up the required environment:

conda env create -f environment.yml

Data Preprocessing:

To preprocess your own data, first navigate to the /data directory:

cd /data

Then, run the respective Python files for different datasets. Note: You'll need to manually change the file name in the codes.

GUIDE-seq:

• On Windows:

python guide_preprocess.py

• On Linux:

python3 guide_preprocess.py

CIRCLE-seq:

• On Windows:

python CIRCLE_process.py

• On Linux:

python3 CIRCLE_process.py

GloVe Embedding:

• On Windows:

python glove_process.py

• On Linux:

python3 glove_process.py

After running the preprocessing scripts, .pkl files with different dimensions will be created for training.

Model Training:

You can modify or add your own models in model_get.py. To train your own model, navigate back to the root directory and start the Jupyter notebook:

jupyter notebook

Then, execute the code in train.ipynb on your local server.

Model Evaluation:

You can either use your trained model for evaluation or use our pre-trained models available in the saved_model folder. Perform the evaluations using cross_validation.ipynb.

Benchmark datasets:

• GUIDE-seq(Hek293t)
• GUIDE-seq(K562)
• CIRCLE-seq
• SITE-seq
• ELEVATION