Abstract

The current landscape of functional data analysis (FDA) predominantly caters to single variable on one view of dataset, overlooking the prevalence of multiview multivariate functional data in biomedical research. While canonical correlation analysis (CCA) stands out as a popular choice for integrative analysis, its applicability is limited to cross-sectional data, failing to address longitudinal or functional data scenarios. In response to these limitations, we propose an innovative integrative sparse functional canonical correlation analysis approach for multiview data. This novel framework aims to tackle the challenges posed by multiview functional datasets, seamlessly integrating both cross-sectional and longitudinal/functional data while accounting for sparsity. The method aims to identify linear combinations of variable functions for each view such that the correlation between the sets of linear combinations is maximized. Our method will also identify interpretable variables that maximize such association over time. We will conduct simulation studies to evaluate the effectiveness of our approach. We will use our method to investigate multi-omics biomarkers during the onset and progression of type 2 diabetes.