A Method for Identifying Essential Proteins Based on Deep Convolutional Neural Network Architecture with Particle Swarm Optimization

The rapid technical advancement of high-throughput sequencing in recent years has accumulated amounts of data representing relationships between protein pairs, which makes it possible to identify essential proteins by extracting features of nodes in Protein-Protein Interaction (PPI) network. Generally speaking, the existing network based computational methods for identifying essential proteins can be divided into sorting and classification, which are typically represented by centrality and machine learning-based methods. Either of the methods mentioned above requires feature engineering, which needs a lot of human experience and priori knowledge. In this case, with the continuous development of deep learning technology, a series of feature-free essential protein identification methods have been proposed to efficiently deal with large volumes of data. However, these methods often take a lot of time to design the network architecture and adjust parameters. In order to solve the limitation of deep learning-based recognition algorithm, in this paper, we propose a novel method base on particle swarm optimization (PSO), which is able to automatically build a deep convolutional neural network (CNN) to identify essential proteins, called psoCEP. The experiments were conducted on S. cerevisiae dataset, and the comparative results show the effectiveness of the new proposed method compared with the sorting and classification methods, which has higher accuracy, F-measure and AUC than the sorting and classification methods.