B1223
Title: Wrapped normal graphical models
Authors: Anna Gottard - University of Firenze (Italy) [presenting]
Agnese Panzera - University of Florence (Italy)
Abstract: Directional distributions are widely used in many research fields such as biology, medicine, geography or meteorology. Most of the studies focus on the univariate or bivariate case, but some interesting fields require higher dimensional settings. In this framework, the protein structure prediction problem is considered of great interest. Graphical models are a widely used class of multivariate models, where the conditional independence structure of a set of variables can be summarised by a graph. The random variables are represented by the nodes in the graph and conditional independence by missing edges. We introduce a general theory for graphical models for the multivariate Wrapped Normal distribution that allows studying the conditional independence structure of the dihedral angles of each amino acid of a protein. Wrapped Normal graphical models inherit most of the properties of the ordinary Gaussian graphical models, such as closeness to conditioning and marginalisation, decomposability and so on. We provide an interesting interpretation of model parameters and discuss inferential issues.
