seqCrispr

Overview

This project is intended to generate a model for Crispr-Cas9 targeting efficiency prediction.

Below is the layout of the whole model.

This model includes four components:

• embedding layer
• convolutional neural network and recurrent neural network layer
• fully connected layer
• input perturbation layer.

Requirement

• keras
• tensorflow
• h2o
• sklearn
• pandas
• numpy

Usage

Load model

change the model directory in config.py when old model need to be loaded for testing or transfer learning

transfer_learning = True
loaded_model_path = os.path.join(cur_dir, "dataset/best_model/<cellline>_lstm_model.h5")

When no model needs to be loaded, change to

transfer_learning = False

Model training

Test old models only, change "training" in config.py

training = False

Training new models

training = True

Run the program and get prediction result

Make sure the data is in dataset/<cellline>/ folder and execute

./run.sh dataset/<cellline>