Inferring gene regulatory networks with a scale-free property based informative prior

Abstract

Constructing gene regulatory networks (GRNs) with microarray gene data is an essential and challenging task, especially when the underlying structures of networks are not observable in an experimental context. The paper proposes a boosting regression algorithm, called informative prior based GRN construction (ipGRN), to perform GRN inference. The ipGRN utilizes a scale-free based informative prior as well as Bayesian criterion measure to improve inference accuracy. In comparison with three existing methods (NIMOO, lasso and NIR), the ipGRN exhibits a significant improvement of computational accuracy and effectiveness on experiments of synthetic and real datasets. Furthermore, the method was applied to breast cancer data to reconstruct a sub-network of cancer susceptibility genes and achieved better inference results in detecting cancer associated genes.