Biomarkers.

Area of Science:

• Neuroscience
• Biomarker Discovery
• Machine Learning in Healthcare

Background:

• Dementia affects over 50 million globally, necessitating early diagnosis for effective management.
• Current diagnosis often occurs late, after irreversible brain damage.
• Machine learning and multimodal data offer potential for preclinical dementia risk detection.

Purpose of the Study:

• To identify cost-effective, non-invasive dementia risk signatures.
• To develop blood-based biomarkers for preclinical dementia risk prediction.
• To leverage machine learning for early dementia detection.

Main Methods:

• Utilized UK Biobank data with proteomic and metabolomic information.
• Applied machine learning (XGBoost) to predict all-cause dementia.
• Analyzed 3,241 molecules in 872 dementia cases and 29,006 controls over 15.2 years.

Main Results:

• XGBoost achieved ROC AUC scores of 0.85+ for overall and future dementia prediction.
• Identified promising plasma biomarkers for preclinical and early dementia diagnosis.
• Validated existing and discovered novel blood-based dementia biomarkers.

Conclusions:

• Machine learning is effective for preclinical dementia risk prediction.
• Identified crucial proteins for individualized dementia risk assessment.
• Further research will explore dementia subtypes, ethnic, and gender differences in risk profiles.