Deep learning analysis of UPLC-MS/MS-based metabolomics data to predict Alzheimer's disease

Area of Science:

• Neuroscience
• Biochemistry
• Computational Biology

Background:

• Alzheimer's disease (AD) progression can be informed by metabolic biomarkers.
• Developing accurate predictive tools for AD is crucial for timely intervention.
• Ultra Performance Liquid Chromatography Mass Spectrometry (UPLC-MS/MS) offers a powerful platform for metabolomic analysis.

Purpose of the Study:

• To identify and characterize novel diagnostic biomarkers for Alzheimer's disease (AD).
• To develop deep learning (DL) models for predicting AD using metabolomics data.
• To leverage UPLC-MS/MS technology for enhanced AD biomarker discovery.

Main Methods:

• Utilized metabolomics data from 177 participants (78 AD, 99 cognitive normal) from the Alzheimer's Disease Neuroimaging Initiative (ADNI) cohort.
• Applied Least Absolute Shrinkage and Selection Operator (LASSO) for feature selection among 150 metabolomic biomarkers.
• Developed multilayer feedforward neural networks using the H2O DL function for AD prediction.

Main Results:

• Identified 21 key metabolic biomarkers using LASSO, including those involved in glucose and lipid metabolism, particularly bile acid metabolites.
• The best performing DL model achieved an accuracy of 0.881, F1-score of 0.892, and AUC of 0.873.
• Selected biomarkers correlated with APOE-ε4 allele, established AD clinical biomarkers (Aβ42, tTau, pTau), cognitive scores (ADAS13, MMSE), and hippocampus volume.

Conclusions:

• A novel set of diagnostic metabolomic biomarkers for AD prediction was successfully identified.
• These biomarkers are instrumental in developing advanced DL tools for AD.
• The findings support the potential for early AD diagnosis, prognostic risk stratification, and targeted therapeutic interventions.