Mixture Partial Linear Models for High-Dimensional Responses

• Activity Number: 170
• Type: Topic Contributed
• Date/Time: Monday, August 10, 2015 : 10:30 AM to 12:20 PM
• Sponsor: ENAR
• Author(s): Kehui Chen* and Kai Hwang and Michael Hallquist and Beatriz Luna
• Companies: University of Pittsburgh and University of Pittsburgh and University of Pittsburgh and University of Pittsburgh
• Keywords: nonparametric ; smoothing ; resting state fMRI ; functional connectivity

Abstract

This paper is motivated by studies in translational neuroscience, where a common interest is to understand how brain functional measures develops with age, and interacts with covariates such as behavioral measures, cognitive function and mental disorder status. We propose a mixture partial linear model to study the influence of a subject level covariate X on a high dimensional response vector. First, the effect of X on Y is modeled through a mixture of linear functions, and second, the remaining pattern is modeled non-parametrically to allow flexible variations in the response vector. All the model components can be estimated through a difference based procedure. We applied our method to a resting state fMRI data set and found that the brain functional connectivity in basal ganglia shows a developmental local-to-distributed trend through adolescence and the trend is positively associated with IQ values.