A neuro-evolution approach to infer a Boolean network from time-series gene expressions

Abstract

Summary In systems biology, it is challenging to accurately infer a regulatory network from time-series gene expression data, and a variety of methods have been proposed. Most of them were computationally inefficient in inferring very large networks, though, because of the increasing number of candidate regulatory genes. Although a recent approach called GABNI (genetic algorithm-based Boolean network inference) was presented to resolve this problem using a genetic algorithm, there is room for performance improvement because it employed a limited representation model of regulatory functions. In this regard, we devised a novel genetic algorithm combined with a neural network for the Boolean network inference, where a neural network is used to represent the regulatory function instead of an incomplete Boolean truth table used in the GABNI. In addition, our new method extended the range of the time-step lag parameter value between the regulatory and the target genes for more flexible representation of the regulatory function. Extensive simulations with the gene expression datasets of the artificial and real networks were conducted to compare our method with five well-known existing methods including GABNI. Our proposed method significantly outperformed them in terms of both structural and dynamics accuracy. Conclusion Our method can be a promising tool to infer a large-scale Boolean regulatory network from time-series gene expression data. Availability and implementation The source code is freely available at https://github.com/kwon-uou/NNBNI. Supplementary information Supplementary data are available at Bioinformatics online.

Review and assessment of Boolean approaches for inference of gene regulatory networks

“…For example, Vera-Licona et al [32] incorporated evolutionary algorithm as an optimisation procedure to infer Boolean functions represented as Boolean polynomial dynamical systems. NNBNI (Neural Network-based Boolean Network Inference) [33] combines mutual information feature selection, genetic algorithms as a global search technique, and a neural network to represent a regulatory rule. Similarly, RFBFE (Random Forest Best-Fit Extension) [31] employs random forest-based feature selection and Best-Fit Extension to infer large Boolean networks.…”