Predicting Protein-Protein Interactions from Amino Acid Sequences Using SaE-ELM Combined with Continuous Wavelet Descriptor and PseAA Composition

Abstract

Protein-protein interactions (PPIs) are known for its crucial role in almost all cellular processes. Although many innovative techniques for detecting PPIs have been developed, these methods are still both time-consuming and costly. Therefore, it is significant to develop computational approaches for predicting PPIs. In this paper, we propose a novel method to identify new PPIs in ways of self-adaptive evolutionary extreme learning machine (SaE-ELM) combined with a novel representation using continuous wavelet (CW) transform and Chou’s pseudo amino acid feature vector. We apply Meyer continuous wavelet transform to extracting wavelet power spectrums from a protein sequence representing a protein as an image, which allows us to use well-known image texture descriptors for extracting protein features. Chou’s pseudo-amino-acid composition (PseAAC) expands the simple amino-acid composition (AAC) by retaining information embedded in protein sequence. SaE-ELM, a variant of extreme learning machine (ELM), optimizes the single hidden layer feedforward network (SLFN) hidden node parameters using self-adaptive different evolution algorithms. When performed on the PPI data of yeast, the proposed method achieved 87.87 % prediction accuracy with 91.19 % sensitivity at the precision of 82.62 %. Extensive experiments are performed to compare our method with the method base on state-of-the-art classifier, support vector machine (SVM). It is observed from the achieved results that the proposed method is very promising for predicting PPI.