Title: Extensions of graphical models with applications in genetics and genomics
Authors: Pariya Behrouzi - Wageningen University and Research (Netherlands) [presenting]
Ernst Wit - University of Groningen (Netherlands)
Abstract: Several problems related to modeling complex systems are addressed. Fields such as systems genetics, systems biology, epidemiology, and bioinformatics often involve large-scale models in which thousands of components are linked in complex ways. What is perhaps most distinctive about the graphical model approach is its suitability in formulating probabilistic models of complex phenomena in applied fields, while maintaining control over the computational cost associated with these models. In real world, not all datasets are continuous. Discrete data or mixed discrete-and-continuous datasets routinely arise in above-mentioned fields. We introduce a method for reconstructing a conditional independence network from non-Gaussian data, in particular for ordinal and for mixed ordinal-and-continuous data. Such data are common in systems genetics, where the main focus is to understand the flow of biological information that underlies complex traits. We focus on the trait survival: we aim to find loci --locations on a genome-- that do not segregate independently conditional on other loci. The network estimation relies on penalized Gaussian copula graphical models; this accounts for a large number of markers $p$ and a small number of individuals $n$.