Statistical Modeling of Alzheimer's Disease Progression Integrating Brain Imaging and -Omics Data

点击翻译按钮获取中文摘要

Understanding of the etiology of Alzheimer's Disease (AD) is complicated due to the existence of dysregulations at different biological scales, ranging from genetic mutations to structural and functional brain alterations. Most models for studying AD are primarily focused on unimodal analysis, but there is a lack of systematic approaches that can integrate data across multiple scales to study the longitudinal disease progression. For example, the molecular mechanisms of brain atrophy related to progression to AD is not well understood. Although the promise of integrative analysis across multiple scales is increasingly recognized, there has been limited progress in developing interpretable and systematic approaches due the fact that the neuroimaging and -omics features have unique patterns of dependence and it is not immediately clear how to combine these two modalities for modeling progression to AD. Another limitation is that most of the existing methods have focused on delineating biological causes for differences between disease specific phenotypes that does not account for heterogeneity and does not treat the disorder as a continuum, which is recommended as per current NIA guidelines. To address these critical challenges, we develop a suite of statistical methods for modeling disease progression in AD involving longitudinal neuroimaging (MRI) scans and cognitive scores, combined with baseline -omics features and demographic and clinical data. Our integrative longitudinal analysis addresses critical gaps in literature and generates more robust results that are generalizable to more inclusive populations and yields more power in detecting true signals. We use spatially distributed voxel-wise brain surface features derived from MRI scans that provides high resolution interpretations about the changes in brain shape associated with disease progression. We develop predictive models which treats AD as a continuum while integrating data across disease stages and multiple visits in a systematic manner that is able to account for heterogeneity between and within disease stages and provides interpretable insights into longitudinal neuroimaging and baseline -omics features that drive cognition. Our methods can be used for developing individualized prediction trajectories for disease progression, identify latent states that are prognostic for specific disease stages, and predict cognition at future visits that can be directly used for early detection of high-risk individuals. We will develop and train our models using longitudinal ADNI data involving several thousand individuals and validate our findings on an independent longitudinal B-SHARP dataset. The statistical tools and algorithms developed will be made widely available to the broader research community. To our knowledge, our project is one of the first to develop an integrative and interpretable statistical framework for studying the trajectory of disease progression in AD using longitudinal and heterogeneous biomarker data from multiple scales, which provides valuable computational tools for early detection in AD that is of tremendous clinical importance in delivering patent centric outcomes in precision medicine.