Function Approximation Approach to the Inference of Neural Network Models of Genetic Networks

Abstract

A model based on a set of differential equations can effectively capture various dynamics. This type of model is therefore ideal for describing genetic networks. Several genetic network inference algorithms based on models of this type have been proposed. Most of these inference methods use models based on a set of differential equations of the fixed form to describe genetic networks. In this study, we propose a new method for the inference of genetic networks. To describe genetic networks, the proposed method does not use models of the fixed form, but uses neural network models. In order to interpret obtained neural network models, we also propose a method based on sensitivity analysis. The effectiveness of the proposed methods is verified through a series of artificial genetic network inference problems.

The inference method of the gene regulatory network with a majority rule

“…The objective of the gene regulatory network inference is to determine the regulatory interactions among genes in the network. Kimura et al had proposed the inference method using the neural networks [13]. In this study, we use this conventional method as a basic method of our proposed method.…”

“…The differential equation models describe the dynamical characteristics of gene expressions based on the information of correlations among genes. The differential equation models are considered as one of the most outstanding tools to describe a mechanism of mutual interactions of genes, metabolism and so on [9][10][11][12][13][14][15][16][17][18].…”

“…On the other hand, the heuristic method is considered as one of the effective methods for the function approximation and some conventional studies use it for the inference of the gene regulatory network that is defined by the differential equations [12,13]. In these heuristic methods, the method using the neural networks [13] is one of the successful methods, where the neural network is known as an outstanding solver for the function approximation.…”

“…In these heuristic methods, the method using the neural networks [13] is one of the successful methods, where the neural network is known as an outstanding solver for the function approximation. It can approximate unknown functions using the learning algorithm such as the error back propagation.…”

“…That is, the method using the neural networks does not require the particular fixed form of the differential equations. Also, it had been reported that the method can reduce the computational time [13].…”