Abstract
BACKGROUND: Recent studies indicate that gene expression levels in blood may be able to differentiate subjects with Alzheimer's disease (AD) from normal elderly controls and mild cognitively impaired (MCI) subjects. However, there is limited replicability at the single marker level. A pathway-based interpretation of gene expression may prove more robust. OBJECTIVES: This study aimed to investigate whether a case/control classification model built on pathway level data was more robust than a gene level model and may consequently perform better in test data. The study used two batches of gene expression data from the AddNeuroMed (ANM) and Dementia Case Registry (DCR) cohorts. METHODS: Our study used Illumina Human HT-12 Expression BeadChips to collect gene expression from blood samples. Random forest modeling with recursive feature elimination was used to predict case/control status. Age and APOE ɛ4 status were used as covariates for all analysis. RESULTS: Gene and pathway level models performed similarly to each other and to a model based on demographic information only. CONCLUSIONS: Any potential increase in concordance from the novel pathway level approach used here has not lead to a greater predictive ability in these datasets. However, we have only tested one method for creating pathway level scores. Further, we have been able to benchmark pathways against genes in datasets that had been extensively harmonized. Further work should focus on the use of alternative methods for creating pathway level scores, in particular those that incorporate pathway topology, and the use of an endophenotype based approach.