B0487
Title: Latent multimodal functional graphical model estimation
Authors: Mladen Kolar - University of Chicago (United States) [presenting]
Abstract: Joint multimodal functional data acquisition, where data multiple modes of functional data are measured from the same subject simultaneously, has emerged as an exciting modern approach enabled by recent engineering breakthroughs in the neurological and biological sciences. One prominent motivation for acquiring such data is to enable new discoveries of the underlying connectivity by combining signals from multiple modalities. Yet, despite scientific interest in this problem, there remains a gap in principled statistical methods for estimating the graph underlying joint multimodal functional data. To this end, we propose a new integrative framework to estimate a single latent graphical model from multimodal functional data. We take the generative perspective by modeling the data generation process, from which we identify the inverse mapping from the observation space to the latent space as transformation operators. We then develop an estimator that simultaneously estimates the transformation operators and the latent graph via functional neighborhood regression. The approach is motivated and applied to analyzing simultaneously acquired multimodal brain imaging data where the graph indicates the underlying brain functional connectivity.