Document Type
Honors Thesis
Abstract
Alzheimer’s pathology begins years before clinical symptoms, starting early with amyloid-β accumulation, followed by tau deposition and neurodegeneration. Although existing tools, such as cerebrospinal fluid (CSF) screening and PET/MRI imaging, can accurately track disease progression, they are invasive, expensive, and not scalable for population-level screening. So, there is a growing interest in using blood-based plasma biomarkers as a scalable alternative. While recent studies demonstrate strong predictive performance with plasma biomarkers, most rely on a small set of canonical blood-based protein biomarkers such as amyloid-β, p-tau, and neurofilament light. These biomarkers primarily reflect downstream brain pathology and may fail to capture earlier systemic biological alterations associated with whole-body dysregulation that contribute to disease onset and progression.
In this study, we developed an interpretable machine learning framework to identify plasma proteomic signatures capturing systemic biological alterations across the full Alzheimer’s Disease continuum (CN, MCI, AD). Using ADNI proteomics data, we analyzed 146 plasma analytes ranging from inflammatory, growth factor, lipid metabolism, metabolic, and vascular processes in 566 matched subjects through a nested 5×5 cross-validation machine learning framework, incorporating covariate residualization to remove age, sex, and APOE ε4-associated variance. Five machine learning models were evaluated, with SVM achieving the highest performance for adjacent-stage classifications (CN vs MCI: AUC = 0.956 ± 0.017, MCI vs AD: AUC = 0.963 ± 0.011), and Random Forest performing the best for CN vs AD (AUC = 0.836 ± 0.085). SHAP analysis revealed stage-dependent, biologically meaningful patterns: Early stage (CN vs MCI) was characterized by lipid metabolism and immune signaling proteins, later stage (MCI vs AD) was driven by inflammatory, growth factor, and vascular processes, whereas CN vs AD classification reflected a broader systemic dysregulation.
These findings suggest that plasma proteomics can capture broad systemic biological signals associated with Alzheimer’s disease progression. Future work can focus on mapping our machine-learning-identified plasma proteins to multimodal brain measures (MRI, CSF, PET) to better understand how systemic biological signals relate to downstream brain pathology. Overall, our study supports the potential of a scalable, data-driven approach for stage-specific AD characterization.
Disciplines
Bioinformatics | Biomedical Engineering and Bioengineering | Neuroscience and Neurobiology
Publication Date
7-2026
Language
English
Faculty Mentor of Honors Project
Xi Zhu
License
This work is licensed under a Creative Commons Attribution 4.0 International License.
Recommended Citation
Laugi, Hemshankar, "Interpretable Machine Learning of Plasma Proteomics Reveals Stage-Specific Signatures Across the Alzheimer's Disease Continuum" (2026). 2026 Spring Honors Capstones Projects. 14.
https://mavmatrix.uta.edu/honors_spring2026/14
Included in
Bioinformatics Commons, Biomedical Engineering and Bioengineering Commons, Neuroscience and Neurobiology Commons
Comments
I would like to express my sincere gratitude to Dr. Xi Zhu for her exceptional mentorship, encouragement, and support throughout this research. Her constructive feedback at every stage of the project challenged me to think critically and helped me grow as an independent researcher. I would also like to thank Mohammadreza Hosseinzadehketilateh for his critical insights and assistance with this project.