Cross study transcriptomic investigation of Alzheimer’s brain tissue discoveries and limitations

Abstract

Abstract Developing effective treatments for Alzheimer’s disease (AD) likely requires a deep understanding of molecular mechanisms. Integration of transcriptomic datasets and developing innovative computational analyses may yield novel molecular targets with broad applicability. The motivation for this study was conceived from two main observations: (a) most transcriptomic analyses of AD data consider univariate differential expression analysis, and (b) insights are often not transferable across studies. We designed a machine learning-based framework that can elucidate interpretable multivariate relationships from multiple human AD studies to discover robust transcriptomic AD biomarkers transferable across multiple studies. Our analysis of three human hippocampus datasets revealed multiple robust synergistic associations from unrelated pathways along with inconsistencies of gene associations across different studies. Our study underscores the utility of developing AI-assisted next-gen metrics for integration, robustness, and generalization and also highlights the potential benefit of elucidating molecular mechanisms and pathways that are important in targeting a single population.