An Information and Combinatorial Theories-based Supervised Learning Framework for Integrative Inference and Analysis of Genetic Regulatory Networks

A supervised learning framework based on information and combinatorial theories is introduced for the inference and analysis of genetic regulatory networks. First, an associativity measure is proposed to quantify the regulatory strength. Next, a phase-shift metric is defined for detecting regulatory orientations among network components. Thus, this framework can solve undi-rected problems from most current linear/nonlinear relevance methods. For computational redun-dancy, the size of the classified pair candidates is constrained within a multiobjective combinato-rial optimization problem. In comparison with previously reported methods, our flexible approach can be used to efficiently identify a directed biological network that is verified by both synthetic and real-world microarray datasets having different statistical characteristics. Thus, the underlying network-designing mechanisms are deciphered by qualitative and quantitative means