Integrative modeling and dynamic prediction of Alzheimer's disease

 
   Lead Investigator:    Sheng Luo
   Institution      :    Duke University
   E-Mail           :    sheng.luo@duke.edu
   Proposal ID      :    1332



   Proposal Description:
   Impairment caused by Alzheimer???s disease (AD) affects multiple domains (e.g.,   cognition, behavior, and quality of life) and progresses heterogeneously in       time and across domains and individuals. The heterogeneous nature and unknown     pathogenic mechanisms of AD make it impossible to rely on a single outcome to     reflect disease severity and progression. Consequently, AD studies collect data   from multiple sources: longitudinal clinical data (e.g., neuropsychological,      functional, and behavioral assessments), neuroimaging data (e.g., magnetic        resonance imaging, MRI), and genetic data such as single-nucleotide               polymorphisms (SNPs) from genome-wide association studies (GWAS). These           multi-modal data (clinical, neuroimaging, and genetic) are predictive of AD       onset and progression. The growing, expensive public burden of AD has raised      the urgency to efficiently analyze the multi-modal data and to provide accurate personalized prediction. The ultimate goal is to provide targeted prognosis and   disease-modifying therapeutic intervention.                                            Most existing AD models have four well-known limitations: (1) they only      use the baseline measurements or a single longitudinal variable and fail to       utilize the whole longitudinal trajectories of multiple clinical variables (2) they are based on the volume change of brain regions of interest (ROI) derived    from MRI, rather than the more recent and informative high-dimensional            vertex-based morphology analysis (based on the longitudinal changes of cortical thickness in thousands of vertices) (3) they only include a few risk loci from the GWAS data while ignoring the large polygenic contribution from many common    variants of small effect to the overall heritable risk of AD (4) most             importantly, they analyze at a single data source level and do not capture the correlated structure of the multi-modal data