Machine Learning Approach to Select Small Compounds in Plasma as Predictors of Alzheimer's Disease

Area of Science:

• Biochemistry
• Computational Biology
• Neuroscience

Background:

• Alzheimer's disease (AD) poses a significant global health challenge.
• Early and accurate diagnosis of AD is crucial for effective management and therapeutic interventions.
• Current diagnostic methods have limitations, necessitating novel biomarkers.

Purpose of the Study:

• To develop a machine learning-based predictive model for Alzheimer's disease using plasma metabolomics data.
• To identify a specific signature of small molecules capable of distinguishing AD patients from healthy controls.
• To explore the potential of metabolomics as a non-invasive diagnostic tool for AD.

Main Methods:

• Utilized metabolomics data from plasma samples of 94 AD patients and 62 healthy controls.
• Employed machine learning, including Linear Discriminant Analysis (LDA) with Leave-One-Out Cross-Validation.
• Preprocessed and normalized metabolite data, identifying concomitant metabolites for enhanced sample comparison.

Main Results:

• Distinct plasma metabolite profiles were observed between AD patients and controls across various biochemical groups.
• Specific metabolites such as carnosine, 5-aminovaleric acid (5-AVA), cholic acid (CA), and indoxyl sulfate (Ind-SO4) showed promise as AD indicators.
• Combinations of four to five metabolites achieved >75% accuracy in classifying AD, with high sensitivity and specificity.

Conclusions:

• Plasma small molecule data, analyzed via machine learning, can serve as effective predictors for Alzheimer's disease.
• This approach offers a novel, potentially non-invasive diagnostic strategy for AD.
• The findings support advancements in personalized medicine for AD management.