Abstract
The principles governing genotype-phenotype relationships are still emerging(1-3), and detailed translational as well as transcriptomic information is required to understand complex phenotypes, such as the pathogenesis of Alzheimer's disease. For this reason, the proteomics of Alzheimer disease (AD) continues to be studied extensively. Although comparisons between data obtained from humans and mouse models have been reported, approaches that specifically address the between-species statistical comparisons are understudied. Our study investigated the performance of two statistical methods for identification of proteins and biological pathways associated with Alzheimer's disease for cross-species comparisons, taking specific data analysis challenges into account, including collinearity, dimensionality reduction and cross-species protein matching. We used a human dataset from a well-characterized cohort followed for over 22 years with proteomic data available. For the mouse model, we generated proteomic data from whole brains of CVN-AD and matching control mouse models. We used these analyses to determine the reliability of a mouse model to forecast significant proteomic-based pathological changes in the brain that may mimic pathology in human Alzheimer's disease. Compared with LASSO regression, partial least squares discriminant analysis provided better statistical performance for the proteomics analysis. The major biological finding of the study was that extracellular matrix proteins and integrin-related pathways were dysregulated in both the human and mouse data. This approach may help inform the development of mouse models that are more relevant to the study of human late-onset Alzheimer's disease.