A framework for identifying essential proteins with hybridizing deep neural network and ordinary least squares

Essential proteins are indispensable for the maintenance of life activities and play an important role in biological processes. Identification of essential proteins is of great importance for understanding the minimum requirements of cell life, discovering pathogenic genes, drug targets, diagnosis of diseases, mechanism of biological evolution, etc. The state-of-the-art methods suggest that integrating protein-protein interaction networks (PPI) and related features of biological sequences can improve the identification accuracy and robustness of essential proteins. In this paper, a novel method called IYEPDNN is proposed to identify essential proteins based on multidimensional biological attribute information and topological properties of PPI by hybridizing deep neural network (DNN) and ordinary least squares. In IYEPDNN, the gene expression profile, PPI and orthology are integrated together as input features to reduce computational complexity and improve accuracy of DNN. In addition, the ordinary least squares is used to supplement the absent data in the Yeast data set to improve the robustness. Experimental results show that the accuracy of IYEPDNN is 84%, and the number of essential proteins identified by IYEPDNN is much higher than that of the state-of-the-art methods (WDC, PeC, OGN, ETBUPPI, RWAMVL, and so on). The results of this paper show that the correlation between features is the key to improve the accuracy of essential protein prediction, and the correct selection of training data can effectively avoid the problem of training data imbalance in essential protein identification.