Workshop SDS: Registration
View Submission - SDS2022
Title: Bayesian networks for dihedral angles Authors:  Anna Gottard - University of Firenze (Italy) [presenting]
Agnese Panzera - University of Florence (Italy)
Abstract: A crucial topic in structural bioinformatics is predicting the three-dimensional structure of a protein, as determined by dihedral angles. It is believed that the amino acid sequence of a protein, its primary structure, incorporates the information needed to determine its shape, in turn governing the biological activity. For some proteins, the secondary structure and the functionality may vary with the membrane composition of the peptide habitat. A step forward in protein prediction could be understanding the conditional independence structure linking the angles of a protein. Graphical models are a well-known tool for analysing conditional independence between random variables. Dihedral angles are a special kind of random variable called a circular variable. The intrinsic characteristics of this kind of variable require ad hoc distributions. We explore possible specifications of Bayesian Networks for dihedral angles, assuming that the primary structure provides a natural ordering of the nodes. We analyse alternative parameterisations of Bayesian Networks for Conditional von Mises distribution and Inverse Stereographic distribution. We also provide possible inferential procedures for estimation and graph learning. As an illustration, we apply the proposals to the Methionine data.