Development of Plasma Protein Classification Models for Alzheimer's Disease Using Multiple Machine Learning...

Area of Science:

• Neuroscience
• Biomarker Discovery
• Computational Biology

Background:

• Alzheimer's Disease (AD) diagnosis relies on invasive cerebrospinal fluid (CSF) biomarkers, causing patient discomfort.
• Limitations in current detection methods present challenges for effective AD management.
• Plasma-based biomarkers offer a less invasive, more cost-effective diagnostic alternative.

Purpose of the Study:

• To develop and validate machine learning models for AD detection using plasma proteomic data.
• To identify novel plasma protein biomarkers associated with Alzheimer's Disease.
• To explore the relevance of identified biomarkers in AD pathogenesis and aging.

Main Methods:

• Utilized a dataset of 120 plasma proteins from AD patients and cognitively normal individuals.
• Applied diverse machine learning algorithms (EBlasso, EBEN, XGBoost, LightGBM, TabNet, TabPFN) for classification.
• Performed gene ontology, pathway enrichment, and literature review to assess biomarker relevance.

Main Results:

• Machine learning models achieved high diagnostic performance (AUROC and accuracy >0.9).
• Consistently identified predictor proteins included Angiopoietin-2 (ANG-2), EGF, IL-1α, and PDGF-BB, with established links to AD.
• The identified biomarker pool was significantly enriched with aging-related proteins (p=0.040).

Conclusions:

• Cutting-edge algorithms enhance the development of plasma-based AD prediction models.
• The identified proteins may serve as novel therapeutic or preventative targets for Alzheimer's Disease.
• External validation in diverse populations is crucial to confirm the generalizability of these findings.