In a recent study published in Science Advances, researchers from King’s College London have explored the potential of metabolomic aging clocks, developed using machine learning models, to predict health outcomes and life span. This study aimed to benchmark the accuracy, robustness, and relevance of metabolomic aging clocks in assessing biological aging, offering new insights into aging beyond chronological age.
Background
While chronological age is a fundamental measure of time passed since birth, it does not capture the physiological variances observed in aging individuals. Biological aging, which reflects cellular and molecular damage over time, is linked to health outcomes and susceptibility to chronic diseases. Advances in omics technologies, such as metabolomics, have enabled researchers to study aging by profiling small molecules, or metabolites, which are products of metabolic pathways. These metabolites have been associated with aging-related diseases, making them ideal candidates for studying aging mechanisms and predicting future health outcomes.
Recent efforts to construct “aging clocks” from omics data have shown promising results. These clocks estimate biological age by analyzing molecular patterns, such as metabolites, and have been linked to various health outcomes. However, challenges remain in optimizing the predictive power of these models and enhancing their interpretability, particularly in the realm of metabolomics.
The Current Study
The study involved analyzing plasma metabolite data from the U.K. Biobank, which included 225,212 participants aged 37 to 73 years. Researchers used nuclear magnetic resonance (NMR) spectroscopy to assess 168 metabolites, including lipid profiles, amino acids, and glycolysis products. Several exclusion criteria were applied to ensure quality data, including the removal of pregnant women, data inconsistencies, and extreme metabolite values.
Seventeen machine learning algorithms, such as linear regression, tree-based models, and ensemble techniques, were employed to develop metabolomic aging clocks. These models aimed to predict participants’ chronological age based on their plasma metabolite profiles. To validate the models, the researchers applied a nested cross-validation approach, a method that rigorously tests models on various data subsets to reduce overfitting.
The models were evaluated using accuracy metrics like mean absolute error (MAE), root mean square error (RMSE), and correlation coefficients. The Cubist regression model emerged as the most effective, achieving an MAE of 5.31 years—better than other models, including multivariate adaptive regression splines (MAE = 6.36 years). Researchers also adjusted predictions to correct for biases in age predictions, improving their accuracy for both younger and older age groups.
Study Design Overview
The study incorporated a variety of statistical tools and visualizations, including a histogram of the participants’ chronological ages, scatter plots of metabolite levels across different age groups, and hazard ratio analysis linking metabolite levels to all-cause mortality. The hazard ratio analysis demonstrated that certain metabolites had significant associations with both chronological age and mortality risk.
Results
The study revealed that metabolomic aging clocks, derived from plasma metabolite profiles, could effectively differentiate between biological and chronological age. Among the models tested, the Cubist regression model stood out as the most accurate, showing strong associations with health markers and mortality risk.
Positive values for MileAge delta—the difference between predicted and actual age—were linked to accelerated aging, frailty, shorter telomeres, higher morbidity, and an increased risk of mortality. Specifically, for every 1-year increase in MileAge delta, the risk of all-cause mortality increased by 4%, with hazard ratios exceeding 1.5 in extreme cases. Individuals with accelerated aging also reported poorer self-rated health and were more prone to chronic illnesses.
Frailty and telomere attrition showed notable associations with MileAge delta, with some of the differences corresponding to an 18-year discrepancy in frailty index scores. Notably, women exhibited higher MileAge deltas than men across most models.
The study also confirmed the non-linear nature of the relationship between metabolite levels and age. Statistical corrections were crucial for improving model accuracy, particularly for younger and older individuals. The study further demonstrated that metabolomic aging clocks captured unique health-relevant signals that outperformed simpler aging markers, though decelerated aging (negative MileAge delta) did not always correlate with better health outcomes, indicating the complexity of biological aging.
Conclusions
This study demonstrated the effectiveness of metabolomic aging clocks in predicting biological aging and its associated health outcomes. The Cubist rule-based model showed superior predictive accuracy, linking metabolite-derived ages to key health markers and mortality risk. The results suggest that metabolomic aging clocks hold significant potential for proactive health management, risk stratification, and personalized medicine.
The study emphasizes the need for further validation in diverse populations and across longitudinal data to confirm the broader clinical applicability of metabolomic aging clocks. By showcasing how metabolomic profiles can provide actionable insights into aging, the study sets a new benchmark in aging research and highlights the growing potential of metabolomics in understanding and managing aging.
You Might Be Interested In:
-
New Children’s Hospital in Cambridge to Pioneer Integrated Care for Physical and Mental Health
-
Differences in Brain Cell Synapses May Explain Vulnerability to Autism and Schizophrenia
-
Elderberry Juice Linked to Improved Metabolic Health and Weight Management
