The code were tesed on Linux and Mac OS systems.

Note:

• Keras should be run with tensorflow as its backend.
• ViennaRNA, a C code library for prediction of RNA secondary structure, needs to be downloaded before installation.

The required software/packages are:

• python=3.6.5
• numpy=1.14.0
• scipy=1.0.0
• h5py=2.7.1
• tensorflow=1.8.0
• keras=2.1.6
• scikit-learn=0.19.1
• biopython=1.71
• viennarna=2.4.5
• matplotlib
• DotMap
• GPyOpt
• pandas

It is worth noting that when the computing environment(e.g, the version of tensorflow or biopython) changes, the prediction results might change slightly, but the main conclusion won't be affected.

conda create -n crispr python=3.6.5 ipykernel matplotlib pandas numpy=1.14.0 scipy=1.0.0 h5py=2.7.1 tensorflow=1.8.0 keras=2.1.6 scikit-learn=0.19.1 biopython=1.71 viennarna=2.4.5
pip install GPyOpt
pip install DotMap
ipython kernel install --user --name crispr --display-name "Python3(crispr)"

Installation time depends on your own network environment.

Demos were included in the Demo.ipynb file. It contains prediction, metrics and model training demos.

HyperParametersSearching.ipynb, HyperParameters searahcing demonstrates how to get the optimized hyperprameters of DeepHF model.It is time consuming.

• feature_util.py contains the code for extracting position related features and biological features.

• prediction_util.py contains the core code of prediction module for the website www.DeepHF.com.

• training_util.py provides the code for training model in your own computing environment. The optimized hyperparameters is only fit for the aformentioned software/package environment.

• data/wt_seq_data_array.pkl , features and experimental edit efficiency data for WT-SpCas9. It can be used to train the model.

• data/esp_seq_data_array.pkl, features and experimental edit efficiency data for eSpCas9(1.1). It can be used to train the model.

• data/hf_seq_data_array.pkl , features and experimental edit efficiency data for Cas9-HF1. It can be used to train the model.

• models/DeepWt_U6.hd5,the final model file of DeepWt_U6 used in the DeepHF wibsite.

• models/DeepWt_T7.hd5,the final model file of DeepWt_T7 used in the DeepHF wibsite.

• models/esp_rnn_model.hd5,the final model file of eSpCas9(1.1) used in the DeepHF wibsite.

• models/hf_rnn_model.hd5,the final model file of Cas9-HF1 used in the DeepHF wibsite.