Title: Sparse additive graphical models
Authors: Hyonho Chun - Boston University (United States) [presenting]
Abstract: High-throughput technologies frequently appear in genomics, proteomics and metabolites studies, and provide ample opportunities to explore dependence among tens of thousands of biological components. Due to the complexity of biological dependence (e.g. non-linearity and presence of outliers), it is still an active research problem to estimate a compatible dependence structure and to identify subgroups from this dependence structure. We propose sparse additive graphical models by jointly estimating additive components of all variables. By estimating functional components of all variables, the estimated dependence becomes compatible to a probability density function, which is an exciting advancement in multivariate approaches as our approach yields an extra benefit of facilitating missing value imputation.