Inference of gene regulatory networks based on the Light Gradient Boosting Machine

Abstract

Inference of gene regulatory networks (GRNs) is one of the major challenges in molecular biology, understanding of which can reveal the regulatory relationship between transcription factors (TFs) and target genes. Although in the past decades many methods were developed to reconstruct GRNs, the accuracy of traditional methods can be further improved. In this work, we proposed a new method, GRN-LightGBM (Light Gradient Boosting Machine), to reconstruct GRNs. GRN-LightGBM is a non-linear. Ordinary differential equations (ODEs) model established by LightGBM, which is considering regulatory and target genes for a specific gene. Furthermore, GRN-LightGBM utilizes time-series data, steady-state data, and temporal time-delay data together to evaluate the features of regulatory genes important for target genes. GRN-LightGBM is evaluated both in the DREAM4 simulated datasets and Escherichia coli real datasets. The results show that the proposed method outperforms other popular inference algorithms in terms of area under the receiver operating characteristic curve (AUROC) and area under the precision-recall curve (AUPR).