Systemic metabolic disruptions often serve as a silent precursor to the clinical symptoms of Alzheimer's disease (AD). Although central nervous system pathology is defines AD, identifying stable, blood-based biomarkers that map the transition between disease stages remains a key challenge. This study utilized a supervised machine learning framework to differentiate Early Mild Cognitive Impairment from Late Mild Cognitive Impairment (EMCI vs. LMCI) using serum-derived metabolomic data from the Alzheimer's Disease Neuroimaging Initiative (n=643). A two-stage feature selection strategy combining univariate ANOVA F-testing and Recursive Feature Elimination identified a reproducible 15-metabolite panel across randomized data partitions. A Random Forest classifier achieved a mean 5-fold cross-validation accuracy of 81.4% (F1 = 81.1%) during model development and maintained 79.4% accuracy (F1 = 79.5) on an independent testing set. The metabolite panel highlights dual dysfunction in the gut-liver-brain axis and mitochondrial energy homeostasis, specifically featuring bile acid derivatives and fatty acid species. To determine whether these features vary continuously with disease severity, we applied multiple regression models (SGD, PLS, SVR, ElasticNet) to predict cognitive scores (e.g., MMSE, ADAS13) and neuroimaging volumes (e.g., Hippocampus, Entorhinal). Across all outcomes, predictive performance was uniformly low (R2 < 0.1), indicating that metabolic disruptions are more predictive of disease state (EMCI vs. LMCI) than of quantities associated with clinical and anatomical changes. Collectively, these findings indicate that serum-derived metabolic profiles may distinguish discrete transitions between MCI stages. Such biomarkers may provide a sensitive, accessible tool for clinical staging, identifying the onset of late-stage impairment before structural or psychometric decline is apparent.
Acknowledgements: This research was supported by the Robert Smith Summer Research Fellowship. I would like to thank Dr. Chuah for their guidance and mentorship throughout this project.