Translatability of Animal Models for Alzheimer's Disease Using a Machine Learning Based Workflow

Despite significant investment, no effective disease-modifying therapies for Alzheimer's disease (AD) have been developed to date. As understanding of the underlying causes of AD evolves, numerous animal models have been generated to study the disease. However, persistent therapeutic failures raise questions about the reasons for these shortcomings, including whether they stem from poor target selection and/or limitations in replicating key aspects of AD pathophysiology in animal models. In this study, a machine learning-based workflow previously reported in the literature was modified and used to identify shared dysregulation in phenotype-defining pathways across both animal models and human datasets—termed translatable pathways. This approach provided a framework for assessing the translational relevance of three widely used AD models: APP/PS1, 3×Tg, and 5×FAD, from hippocampal microarray data. The analysis suggested no translatable pathways in the APP/PS1 and 3×Tg preclinical models, whereas key pathways were identified in the 5×FAD (SREBP control of lipid synthesis and cytotoxic T-lymphocyte pathways) model. Additionally, applying the workflow to publicly available microarray data from ibuprofen-treated mice accurately predicted the clinical failure of ibuprofen for treating AD in human trials. This study highlights the importance of evaluating the translatability of animal models to human disease and provides a suitable framework for improving the selection of preclinical models in Alzheimer's research.