Integrating gene expression and phenotypic information to analyze Alzheimer's disease.

The assessment of the relationship between gene expression profiling, clinical and histopathological phenotypes would be better suited to understanding Alzheimer's disease (AD) pathogenesis. We developed a multiple linear regression (MLR) method to simultaneously model three variables - Mini-Mental Status Examination (MMSE) score, neurofibrillary tangles (NFT) score and gene expression profile - to identify significant genes. These genes were also used to distinguish subjects with incipient AD from healthy controls. Finally we investigated the behavior of the significant genes across the entorhinal cortex and hippocampus of AD subjects in two different Braak stages. Results indicate that integrating multiple phenotypic and gene expression information of samples increases the power of methods while analyzing small datasets. The MLR method could identify significant genes at reasonable false discovery rates (FDRs), thereby providing a choice of reasonable FDRs. The accuracy in discriminating between subjects affected and unaffected by AD using MLR identified genes was high. We found that transcription and tumor suppressor responses do begin quite early in AD and therefore should be the target of drugs. Several genes were consistently up/down-regulated across the two brain regions and Braak stages and, therefore, can be used as predictive markers to detect AD at an earlier stage.