Large-Scale metabolomics: Predicting biological age using 10,133 routine untargeted LC-MS measurements

Area of Science:

• Gerontology
• Metabolomics
• Computational Biology

Background:

• Untargeted metabolomics studies biological age, a complex trait influenced by numerous factors.
• Existing geroscience studies often suffer from insufficient sample sizes and potential biases.
• Large-scale metabolomics data from routine toxicologic blood measurements are underutilized due to experimental variability.

Purpose of the Study:

• To analyze biological age using a large dataset of routine blood measurements (~10,000 samples).
• To develop and validate a robust computational method to overcome experimental effects in untargeted metabolomics data.
• To identify known and novel biomarkers associated with biological aging.

Main Methods:

• Analysis of ~10,000 untargeted metabolomics blood samples using ultra-high pressure liquid chromatography-quadruple time of flight mass spectrometry (UHPLC-QTOF).
• Development and application of a custom neural network model for chronological age prediction, compared against existing normalization methods.
• Utilized Shapley Additive exPlanations (SHAP) for feature importance to identify age-related metabolites.

Main Results:

• The custom neural network accurately predicted chronological age with an RMSE of 5.88 years (r² = 0.63), outperforming existing methods.
• Identified known aging markers including kynurenine, indole-3-aldehyde, and acylcarnitines.
• Discovered cyclo(leu-pro) as a potential novel aging biomarker, validating tryptophan and acylcarnitine metabolism's link to aging.

Conclusions:

• Robust computational methods, like neural networks, can effectively analyze large, uncontrolled LC-MS metabolomics datasets for aging research.
• This approach reduces bias and enhances the power of metabolomics studies in geroscience.
• The findings confirm the association of specific metabolic pathways with the aging process and highlight potential new biomarkers.