Model selection in the construction of biological networks under the steady-state conditions

The model selection is a decision problem to choose which variables should be included in a statistical model among all plausible models that could be constructed. There are many applications of this problem in different fields from social to mathematical sciences. Here, we particularly deal with the model selection in the construction of the biological networks when the activation of the systems is described under the steady-state condition. The common features of biological networks are their high dimensions, sparsities and interdependences between networks’ components. Due to these challenges, many model selection criteria such as AIC and BIC cannot be successfully applicable in this field. In this study, as the novelty, we suggest ICOMP (information complexity criterion) and its close alternatives, so-called, CAIC and CAICF, in modelling real biological networks with Gaussian graphical model (GGM) and loopbased multivariate adaptive regression splines (LMARS). GGM is one of the common graphical models in systems biology and LMARS is a recently suggested nonparametric model as an alternate of GGM. In our analyses, we initially derive ICOMP, CAIC and CAICF for these two models and then compare their performances with their major criteria, namely, RIC and STARS in GGM and GCV in LMARS