Biomarkers.

Area of Science:

• Neuroscience
• Biomarker Research
• Machine Learning in Medicine

Background:

• Brain age prediction aids in identifying abnormal aging and early neurological disorders, including Alzheimer's disease (AD).
• Multimodal datasets offer potential for enhanced precision in brain age prediction.

Purpose of the Study:

• To investigate the efficacy of combining serum and plasma biomarkers with feature selection for improved brain age prediction.
• To assess the impact of multimodal data integration on predictive accuracy for neurological disorders.

Main Methods:

• Utilized data from 150 normal controls (serum) and 100 (plasma), with 65 overlapping participants.
• Employed a 10-times repeated 5-fold cross-validation model to evaluate performance and mitigate overfitting.
• Applied feature selection techniques to optimize prediction by integrating serum and plasma biomarkers.

Main Results:

• The "Serum only" model yielded an RMSE of 5.07 (R²=0.746).
• The "Plasma only" model improved performance (RMSE=4.48, R²=0.786).
• Combining "Serum + Plasma" (RMSE=4.12, R²=0.816) and "Serum + Plasma + Feature Elimination" (RMSE=2.77, R²=0.917) demonstrated superior predictive accuracy.

Conclusions:

• Multimodal integration of serum and plasma biomarkers, coupled with feature selection, significantly enhances brain age prediction.
• Machine learning techniques applied to comprehensive Alzheimer's disease datasets show substantial promise for improving diagnostic capabilities.