Gene regulatory network inference from gene expression data based on knowledge matrix and improved rotation forest

Abstract

Research on gene expression, as well as the regulatory interactions governing it, has been considered one of the significant challenges in recent decades. The process of regulating gene expression determines which genes should be turned on and turned off at specific times. Microarray data provide a wealth of information regarding gene expression levels. Due to the large volume of such data, their analysis requires robust computational methods for identifying and analyzing gene regulation networks. However, scholars in this field encounter numerous challenges, including the high number of genes to be considered, the limited number of samples, and the inherently noisy nature of the data. Accordingly, the main purpose of the present study is to propose a modeling approach for inferring gene regulatory networks using regression-based and rotation forest methods. In this method, after initial processing, the data are divided into different clusters using clustering approaches. This categorization is utilized in the next step to extract the knowledge matrix. Subsequently, regulatory interactions governing genes are identified using an improved version of rotation forest based on t-SNE. Furthermore, the proposed approach has been compared with some significant methods commonly used in this discipline to evaluate its efficiency. The analysis results demonstrate that, due to the incorporation of hidden biological knowledge within the dataset during the network modeling process, the proposed model excels in identifying regulatory mechanisms. It not only aligns with top methods in the discipline but can also, at times, outperform them in identifying gene regulatory networks.