Implementation of "predicting in vitro transcription factor binding sites using DNA sequence + shape"
-
To install Keras with Tensorflow backend, please refer to https://keras.io/#installation.
-
Python 2.7
Firstly, using encode.sh script to preprocess DNA sequences and their corresponding shape features.
- Usage:
'pbmdata' denotes the path of storing experimental data, e.g. /yourpath/pbmdata.
bash encode.sh <pbmdata>
Run DeepBind_K or DeepCNN without using DNA shape information
- Usage: you can excute run.sh script directly, in which you should modify the python command accordingly, e.g.:
python train_val_test.py -datadir ./pbmdata/$eachTF/data -run 'noshape' -model 'shallow' -batchsize 300 -k 5 -params 30 --train
The command '-model' can be a choice of {'shollow', 'deep'}, where 'shollow' means DeepBind_K, and 'deep' means DeepCNN.
Run DLBSS(shallow) or DLBSS(deep) using DNA shape information
- Usage: you can excute run.sh script directly, in which you should modify the python command accordingly, e.g.:
python train_val_test_hybrid.py -datadir ./pbmdata/$eachTF/data -run 'shape' -model 'shallow' -batchsize 300 -k 5 -params 30 --train
The command '-run' can be a choice of {'shape', 'MGW', 'ProT', 'Roll', 'HelT'}, where 'shape' means using all shape features, 'MGW' means using MGW shape feature, and so on.
The command '-model' can be a choice of {'shollow', 'deep'}, where 'shollow' means DLBSS(shallow), and 'deep' means 'DLBSS(deep)'.
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