A Novel Hybrid Framework for Reconstructing Gene Regulatory Networks

Abstract

Much effect has been devoted over the past decade to inference of gene regulatory networks (GRNs). However, the previous methods infer GRNs containing large amount of false positive edges, which could result in awful influence on biological analysis. In this study, we present a novel hybrid framework to improve the accuracy of GRN inference. In our method, network topologies from linear and nonlinear ordinary differential equation (ODE) models are integrated. The additive tree models are proposed for identification of linear/nonlinear models. We also propose a new criterion function that sparse and relevant terms are considered while inferring linear and nonlinear models. Benchmark datasets from Dialogue for Reverse Engineering Assessments and Methods challenge and real biological dataset from SOS DNA repair network in Escherichia coli are used to test the validity of our method. Results reveal that our proposed method can improve the prediction accuracy of GRN inference effectively and performs better than other popular methods.