Abstract

Early prediction of progression to Alzheimer’s disease among individuals with mild cognitive impairment remains challenging, especially when using non-invasive markers. We developed a machine learning framework that integrates blood-based molecular and demographic data to distinguish individuals who progressed to Alzheimer’s disease from those who remained stable. We evaluated genomic, methylation, gene expression, lipid, and bile acid data using both early and late integration strategies. Late integration consistently performed better, and the best model achieved 90.7% predictive performance using genomic and lipid features. Interpretation analyses identified reproducible markers across molecular layers, including variants in genes previously linked to neuronal function, methylation changes in biologically relevant loci, and altered ceramide levels. These findings show that combining blood-based multi-omics and demographic data can improve early prediction of disease progression and support the development of interpretable biomarkers for precision diagnostics.