Nonparametric Bayes-based Heterogeneous “Drosophila Melanogaster” Gene Regulatory Network Inference: T-Process Regression

Hiroki Miyashita, Takuma Nakamura, Yasutoshi Ida, Takashi Matsumoto, and Takashi Kaburagi


bioinformatics, nonparametric Bayesian inference, T-Process, dynamic Bayesian network, reversible jump Markov Chain Monte Carlo


Recent research into time-varying network inference for gene expression data mainly assumes that gene regulatory networks have linear interactions. This assumption is straightforward and requires comparatively simple model building. However, in various previous biological studies, gene expression data have been believed to have nonlinear properties in their regulatory interactions. To address this, we adopted a nonparametric Bayesian regression method (e.g. a Gaussian Process) for predicting interactions into a time-varying network to achieve more flexible regression capability. The proposed method was evaluated on Drosophila melanogaster gene data, which has been used as a benchmark in a number of studies. This dataset, which was measured by a microarray test, is known to include noise. To obtain stronger robustness to noisy data, in our algorithm, we employed the T-Process instead of the conventional Gaussian Process. To the best of our knowledge, this is the first algorithm to apply nonparametric Bayesian regression method to a time-varying gene regulatory network problem. Our basic algorithm employed reversible jump Markov Chain Monte Carlo (RJMCMC) for inference of whole network structures. The method can handle the two inference problems: (i) change point detection and (ii) network structure inference simultaneously.

Important Links:

Go Back