Heterogeneous genetics effects and mediation in Alzheimer's disease

Project Summary Down syndrome (DS) is the most common genetic intellectual and developmental disability and has historically been associated with poor life expectancy. However, improvements in medical care have increased the lifespan of individuals with DS to above 60 years. Individuals with DS have an increased risk of Alzheimer’s disease (AD), largely due to triplication of the amyloid precursor protein (APP) gene on chromosome 21, which results in amyloid-beta (Aβ) plaque deposition. Individuals with DS also frequently have co-occurring medical conditions such as sleep apnea, hypothyroidism, and cardiovascular disease, which contribute to pathology that may be directly or indirectly implicated in AD pathology and cognitive decline. Thanks to increasingly comprehensive datasets that include clinical, biospecimen, and imaging data on individuals with DS, it has become feasible and critical to understand the direct and indirect associations between these co-occurring conditions and AD in DS. By utilizing data from what is currently the largest AD biomarker study in DS, Alzheimer’s Biomarker Consortium for Down syndrome (ABC-DS), we propose to develop innovative statistical approaches combining Bayesian inference and flexible machine learning to efficiently explore individual-level heterogeneity in DS. First, we propose a method based on Bayesian additive regression trees (BART) to capture individual-level differences in the effect of APOE genotype on Aβ levels in the presence of missing covariates. Second, we will develop a BART-based heterogeneous causal mediation model to estimate the indirect effects of DS/trisomy 21 on Aβ mediated by cardiovascular disease and moderated by demographics and patient/family medical history. By successfully completing the project, we expect to elucidate the role of cardiovascular disease for AD in DS with the two novel statistical methods proposed in this project.