Identification of Yeast’s Interactome Using Neural Networks

Ur Rehman, Hafeez, Habib, Usman, Ijaz, Umer, Islam, Naveed, Khan, Atta Ur Rehman and Nawaz, Raheel ORCID: https://orcid.org/0000-0001-9588-0052 (2019) Identification of Yeast’s Interactome Using Neural Networks. IEEE Access, 7. pp. 179634-179645.

Preview

Published Version Available under License Creative Commons Attribution. Download (8MB) | Preview

Abstract

An important aspect for designing precise medical therapies is to have an accurate knowledge of protein protein interactions involved in the process. Next generation sequencing technologies for discovering novel genes are fuelling an information explosion that allows researchers to study these molecules in previously unimagined ways. A profound understanding of these biological components promises great leaps for the field of medical sciences, i.e., by designing personalized/preventive medicines, or by curing the life threatening diseases including many types of cancers etc. However, experimental techniques are slow and noisy; and usually report false interactions. These limitations prompted a surge of interest in computational techniques to infer the true interactions. In this paper, we propose and evaluate a Neural Network based approach for deciphering the interactions of Saccharomyces cerevisiae species proteins. The novelty of this approach lies in integrating the evidences (broadly classified as structural and non-structural) in a hybrid fashion. The structure-based evidences include, geometrical features extracted from individual homolog templates, e.g., interacting residues, interfacing residues, binding sites of proteins etc., while the non-structural evidences include, biological process, molecular function, cellular component and motif based similarities. These features are combined using Neural Network based classifier to predict true interactions. The algorithm showed encouraging results, when benchmarked for Saccharomyces cerevisiae’s interactome, retrieved from the STRING database; with an accuracy of 92% for functional association networks, while on protein interaction networks the accuracy remained 83%.