DeepDom is an ab-initio method for protein domain boundary prediction

• Installation has been tested in Linux and Mac OS X with Python 2.7.
• Since the package is written in python 2.7, python 2.7 with the pip tool must be installed first. DeepDom uses the following dependencies: numpy, scipy, pandas, h5py, keras version==2.1.2, tensorflow==1.3.0 You can install these packages first, by the following commands:

pip install pandas
pip install numpy
pip install scipy
pip install h5py
pip install -v keras==2.1.2
pip install tensorflow (or GPU supported tensorflow, refer to https://www.tensorflow.org/install/ for instructions)

• This is the Tensorflow version, you must change the backend to TensorFlow. If you have run Keras at least once, you will find the Keras configuration file at: $HOME/.keras/keras.json If it isn’t there, you can create it. Change the default configuration file into:

{	
    "image_dim_ordering": "th",
    "epsilon": 1e-07,
    "floatx": "float32",
    "backend": "tensorflow"
}

If you want to use GPU, you also need to install CUDA and cuDNN; refer to their websites for instructions. CPU is only suitable for prediction not training.

To train a model, users should have .fasta file for protein sequences and domain boundary annotations on corresponding protein sequences. Please refer to the sample_data_seq.txt and sample_data_label.txt as examples. Then use the "dataprocess.pl" to transformat the data. (run "perl dataprocess.pl -h" to see the helps) After that, the processed data can be used as input for "train.py" to train a model (run "python train.py -h" to see the helps). Note: the requirement of packages that imported in the code need to be met.

perl dataprocess.pl -input_seq sample_data_seq.txt -input_label sample_data_label.txt -output_seq processed_seq.txt -output_label processed_label.txt

python train.py -seqfil processed_seq.txt -labelfile processed_label.txt -model-prefix custom_model.h5

custom_model.h5 is the model generated, users can use this file to predict and can also use our pre-trained model that mentioned in our paper. The pre-trained model was saved in file "foldmodel_bilstmwrapper_4sum200_80_40nr_sliwin.h5".

To predict domain boundary for protein sequences, firstly, users need to transformat the .fasta sequence using "dataprocess.pl" (run "perl dataprocess.pl -h" to see the helps) and using "predict.py" to predict for protein sequences (run "python predict.py -h" to see the helps). Either users' own model or the model we provided can be used for prediction.

perl dataprocess.pl -input_seq sample_data_seq.txt -output_seq processed_seq.txt

python predict.py -input processed_seq.txt -output predict_result.txt

Or users can use the custom model trained (as shown in 1) by their own data to predict.

python predict.py -input processed_seq.txt -output predict_result.txt -model-prefix custom_model.h5

If you find the codes or our method is useful, please cite our paper "DeepDom: Predicting protein domain boundary from sequence alone using stacked bidirectional LSTM".