Application of machine learning to blood-based biomarkers of Alzheimer's disease in Down syndrome.

INTRODUCTION: Blood-based biomarkers can improve Alzheimer's disease (AD) characterization in Down syndrome (DS). This study applied hierarchical clustering and machine learning-based feature selection to identify biomarkers associated with disease progression. METHODS: Cross-sectional blood-based biomarkers were analyzed from 211 DS participants (n = 79 cognitively stable [CS]; n = 72 mild cognitive impairment [MCI]; n = 60 AD dementia [DS-AD]). These included markers of amyloid, tau, neurodegeneration, and inflammation. Clustering grouped biomarkers. Decision trees classified disease stage, and Shapley values identified the strongest predictors of disease stage. RESULTS: The strongest predictors overall were neurofilament light chain (NfL), tau/amyloid beta (Aβ)40, Aβ42/Aβ40, alpha-2-macroglobulin (A2M), and interleukin (IL)-10. Within the CS group, NfL, tau/Aβ40, A2M, and IL-10 were strong predictors. In MCI, Aβ42/Aβ40, NfL, A2M, and IL-10 were strong predictors. In DS-AD, Aβ42/Aβ40, NfL, and tau/Aβ40 were the top predictors. Cluster membership varied based on disease stage. DISCUSSION: These findings reveal evolving biomarker signatures and clustering patterns across cognitive stages, underscoring their potential for disease monitoring.