Abstract:
|
Recent advances in bioinformatics allow the collection of multi-omics data, which help in understanding the potential pathogenic mechanisms for Alzheimer's disease. However, most existing methods cannot deal with high-dimensional multi-omics profiles. We propose a novel joint Bayesian dynamic prediction framework for the integrative analysis of multiple longitudinal neurocognitive markers and multi-omics data. Specifically, we assume that only a few unobserved latent variables capture biological and technical sources of variability of the omics data. The latent variables reduce the dimensionality and account for the high correlations across data modalities. We further introduce a multivariate functional mixed model (MFMM) for feature extraction from multiple longitudinal outcomes. Extensive simulation studies validate the proposed approach. An application to the Alzheimer's Disease Neuroimaging Initiative studies is provided and identifies several new biomarkers.
|