CRPTS: Predicting Transcription Factor Binding Sites using DNA Shape Features Based on Shared Hybrid Deep Learning Architecture

• To install Keras with Tensorflow backend, please refer to https://keras.io/#installation.
• Tensorflow=1.2, keras=2.0
• Python 2.7

Clone the repositopry into your working space.

Download the data from https://bitbucket.org/wenxiu/sequence-shape/get/2159e4ef25be.zip Firstly, using encode.sh script to preprocess DNA sequences and their corresponding shape features.

• Usage: bash encode.sh
• 'pbmdata' denotes the path of storing experimental data, e.g. /yourpath/pbmdata.

• Usage: you can excute run.sh script directly, in which you should modify the python command accordingly, e.g.: python train_val_test_hybrid.py -datadir ./pbmdata/$eachTF/data -run 'shape' -model 'CRPTS' -batchsize 300 -k 5 -params 30 --train
• The command '-run' means 'shape' using four shape features, and the command '-model' can be a choice of {'CRPTS', 'CRPT'}
• Note that you should change the ouput path in the run.sh script, the naming rule is: 'model_' + args.model + '_' + args.run.
• Type the following for details on other optional arguments: python train_val_test_hybrid.py -h

• If you have any problems, please contact SiguoWang: siguo_wang@163.com