Modeling Networks for Prediction and Causal Inference in Quantitative Genetics and Genomics

Abstract Text: Networks can be used to represent biological systems, which are composed of interconnected components. Each component or variable in a network is symbolized by a node (or vertice), while relationships among them are symbolized by edges.  In genetics, for example, networks are used to represent gene regulation systems, co-expression, epistatic interactions, etc. A network modeling approach commonly used in genetic analyses refer to correlation networks. Correlation networks, however, are not able to reveal causal relationships between variables. Conversely, some other network modeling approaches do explore potential direction of edges connecting nodes, by probing conditional independences encapsulated in the joint distribution of the set of variables. Such methods belong to a set of data analysis tools termed Stochastic Graphical Models, which include techniques such as path analysis, Bayesian networks, and structural equation models. In this talk I will present a brief overview of some graphical modeling approaches, and illustrate their application in prediction and causal inference in the context of quantitative genetics and genomics. Examples to be presented involve complex phenotypic traits and genomic information, such as molecular markers and gene expression scores, in different livestock species.

Keywords: Networks, Graphical Models, Causal Inference, Prediction, Markov blanket