Predicting protein–protein interactions from protein sequences by a stacked sparse autoencoder deep neural network

Abstract

Protein-protein interactions (PPIs) play an important role in most of the biological processes. How to correctly and efficiently detect protein interaction is a problem that is worth studying. Although high-throughput technologies provide the possibility to detect large-scale PPIs, these cannot be used to detect whole PPIs, and unreliable data may be generated. To solve this problem, in this study, a novel computational method was proposed to effectively predict the PPIs using the information of a protein sequence. The present method adopts Zernike moments to extract the protein sequence feature from a position specific scoring matrix (PSSM). Then, these extracted features were reconstructed using the stacked autoencoder. Finally, a novel probabilistic classification vector machine (PCVM) classifier was employed to predict the protein-protein interactions. When performed on the PPIs datasets of Yeast and H. pylori, the proposed method could achieve average accuracies of 96.60% and 91.19%, respectively. The promising result shows that the proposed method has a better ability to detect PPIs than other detection methods. The proposed method was also applied to predict PPIs on other species, and promising results were obtained. To evaluate the ability of our method, we compared it with the-state-of-the-art support vector machine (SVM) classifier for the Yeast dataset. The results obtained via multiple experiments prove that our method is powerful, efficient, feasible, and make a great contribution to proteomics research.

Protein Interface Complementarity and Gene Duplication Improve Link Prediction of Protein-Protein Interaction Network

“…Many sequencebased machine learning methods (Huang et al, 2016;An et al, 2017;Wang et al, 2017a,b;You et al, 2017) have been developed. Based on the primary sequences of proteins, they use machine learning algorithms, such as Neural Network (Wang et al, 2017b), Support Vector Machine (SVM) (Wang et al, 2017a), and rotation forest (You et al, 2017) to predict proteinprotein interactions.…”

Lasagna: Multifaceted Protein-Protein Interaction Prediction Based on Siamese Residual RCNN

“…These features measure physicochemical properties of the 20 canonical amino acids, and aim at summarizing full sequence information relevant to PPIs. More recent works [48,51] propose the use of stacked autoencoders (SAE) to refine these heterogeneous features in lowdimensional spaces, which improve the aforementioned models on the binary prediction task. On the contrary, fewer efforts have been made towards multi-class prediction to infer the interaction types [44,64] and the regression task to estimate binding affinity [47,59].…”

Biomarker2vec: Attribute- and Behavior-driven Representation for Multi-type Relationship Prediction between Various Biomarkers

“…Although many attempts have been made to detect the uncovered relationships through various methods including matrix factorization [12], machine learning [13] and network analysis [14]. The incompleteness of the data constrains the credibility of the prediction results accompanied with higher FPR and FNR [15].…”