MAHyNet: Parallel Hybrid Network for RNA-Protein Binding Sites Prediction Based on Multi-Head Attention and Expectation pooling
Introduction
In this work, a new parallel network that integrates the multi-head attention mechanism and the expectation pooling is proposed, named MAHyNet. The left-branch network of this model mixes convolutional neural network and gated recurrent neural network, and its right-branch network is a two-layer convolutional neural network, which can extract the features of one-hot and RNA base physicochemical properties, respectively.
Requirements
- Keras = 2.1.6
- tensorflow-gpu =1.8.0
- h5py
- pool
- tqdm
- sklearn
Non-10-fold cross-validation
python generate_hdf5.py
python generate_hdf5_ph.py
python train_data.py 0 0 53
python save_result.py
10-fold cross-validation
python generate_hdf5_10.py
python generate_hdf5_10ph.py
python train_ten_data.py 0 0 53
python see_10_fold.py