Inference of time series chain graphical model

Biological data can have complex structures due to the high dependence on genes, limited observations, and sparse interactions. This complexity increases when we also consider the influence of time on the construction of the system. This study proposes a comparative study among the penalized likelihood method and two well-known Bayesian approaches under time chain Gaussian copula graphical model. The underlying Bayesian methods are the birth-death Markov chain Monte Carlo (BDMCMC) and reversible jump MCMC (RJMCMC) algorithms. In the implementation of RJCMCMC, we also propose three types of Bayesian schemes, namely, semi-Bayesia, and full-Bayesian approaches, and modified RJMCMC with adaptive turning parameters to estimate model parameters of near-time network structures. In the comparative analyses, we evaluate the performance of the three approaches by using different simulated datasets, and in the assessment, we compute both specificity and Matthew's correlation coefficient while comparing their accuracy.