Multi-modal machine learning and gut microbiome pathway analysis for Alzheimer's risk prediction

Area of Science:

• Neuroscience
• Biostatistics
• Microbiome research

Background:

• Early Alzheimer's disease (AD) detection needs accessible, non-invasive methods beyond traditional biomarkers.
• Machine learning offers potential for analyzing complex datasets to identify early AD risk factors.

Purpose of the Study:

• To develop and validate a multi-modal machine learning framework for early Alzheimer's disease risk assessment.
• To evaluate the predictive performance of various metadata categories (demographic, dietary, lifestyle, nutritional, medical) and microbiome data for AD risk.

Main Methods:

• Analysis of 9832 participants' metadata across five categories, with feature selection using Pearson correlation and chi-squared tests.
• Training four machine learning algorithms with 10-fold cross-validation and SMOTE, validated on 1967 samples.
• 16S rRNA sequencing for microbiome composition analysis in 2000 participants.

Main Results:

• Medical history (AUC=0.871) and dietary patterns (AUC=0.874) demonstrated the highest predictive performance for AD risk.
• Microbiome analysis identified a Prevotella/Bacteroides ratio of 1.921, indicating dysbiosis linked to dietary factors and potential neuroinflammation.
• Performance of demographic, lifestyle, and nutritional domains was significantly lower (p < 0.001).

Conclusions:

• Non-invasive, multi-modal screening combining medical and dietary data shows promise for Alzheimer's disease risk stratification.
• Preliminary evidence suggests gut-brain axis dysbiosis as a potential mechanistic pathway in AD, requiring further validation in larger cohorts.