AUTHOR=Li Feifei , Zhu Fei , Ling Xinghong , Liu Quan 

TITLE=Protein Interaction Network Reconstruction Through Ensemble Deep Learning With Attention Mechanism

JOURNAL=Frontiers in Bioengineering and Biotechnology

VOLUME=Volume 8 - 2020

YEAR=2020

URL=https://www.frontiersin.org/journals/bioengineering-and-biotechnology/articles/10.3389/fbioe.2020.00390

DOI=10.3389/fbioe.2020.00390

ISSN=2296-4185

ABSTRACT=Protein interactions play a fundamental role in the life system and in studying life phenomena with a considerable amount of literature having been published on analyzing and predicting protein interactions, such as support vector machine method, homology-based method, and similarity-based method, each with their own advantages and disadvantages. Studies have revealed that the single method is dissatisfactory in predicting protein interactions, declaring the need for a comprehensive method that combines the advantages of various methods. As ensemble learning, combining multiple models, considers advantages of various methods which offers feasible solutions for strengthening correct prediction and reducing error prediction, serving as a good alternative for comprehensive learning. On the other hand, most existing methods for predicting protein interactions require prior domain knowledge, making it difficult to effectively extract protein features. Deep learning approximates complex function through the nonlinear network structure, which is able to learn the essential features of data sets from a small set of samples and to extract feature representations from various data sets. On this basis, we propose a deep ensemble learning method for predicting protein interactions which encodes protein sequences by LD, AC, CT, PseAAC, and combines the vector representation of each protein in the protein interaction network. Then it takes advantage of multi-layer convolutional neural network to automatically extract protein features and construct attention mechanism to analyze the deep-seated relationship between proteins. We set up four different structures of deep learning models. In the ensemble learning model, we used a five-fold cross validation method to predict the protein interaction network by combining 16 models.  The experiments on 5 independent data sets indicated that our algorithm could learn from different DNNs and representations better than other approaches.