Artificial Intelligence
Using AI to Predict Health and Longevity
A study finds the best AI models to use for assessing biological age and life span.
Posted December 31, 2024 Reviewed by Margaret Foley
Not all artificial intelligence (AI) machine learning algorithms are created equal. Which one is the best for determining human biological age and life span? The answer is an important one, as the adoption rate for digital health and AI tools continues to increase worldwide among researchers, clinicians, and health care providers. Researchers at the Institute of Psychiatry, Psychology & Neuroscience (IoPPN) at King's College London performed an extensive evaluation on a wide range of different AI algorithms to determine which ones were best at predicting biological age from human blood and published their study in Science Advances.
“This study presents a comprehensive comparison of machine learning algorithms for developing metabolomic aging clocks, benchmarking a wide range of models under consistent conditions in one of the largest metabolomics datasets available globally,” wrote IoPPN lead author Dr. Julian Mutz, with co-authors Raquel Iniesta and Cathryn M. Lewis.
In biochemistry, metabolomics is the scientific field that studies the chemical substances produced by an organism, cell, or tissue as a result of metabolism, called metabolites. Metabolism is the sum of all chemical processes at the cellular level that sustain life in organisms. It is a familiar term. People with a high metabolism burn calories at a faster rate during rest and activity than those with a slow metabolism. What might be not so familiar to the average person is metabolism can be further subdivided into two types: destructive metabolism, called catabolism, the processes involved in degradation, or creative metabolism, also known as anabolism, the processes involved in synthesis. Catabolism is the breakdown of complex molecules into simple ones to release or create energy. It includes the processes that convert molecules from food and liquids into smaller units of biomolecules that can be oxidized or used for anabolism. Anabolism, the exact opposite, is the processes that require energy to build complex molecules from simple ones. Pregnancy, bone growth mineralization, wound healing, and muscle mass buildup are all examples of anabolic processes.
“The aim of this study was to compare multiple machine learning algorithms for developing metabolomic aging clocks using nuclear magnetic resonance (NMR) spectroscopy data in the UK Biobank,” wrote the scientists.
To train and validate 17 different AI algorithms for this study, the research team used data spanning 168 different metabolites from the plasma in the blood of over 225,000 participants in the UK Biobank database of middle-aged and older adults with the mean age of 56.97 years old.
The UK Biobank metabolite data was extracted from the blood plasma using nuclear magnetic resonance (NMR) spectroscopy, a noninvasive chemical analysis technique that obtains characteristics of organic molecules by recording the interaction between electrically charged nuclei of atoms exposed to an external magnetic field and radiofrequency waves.
The AI algorithms were evaluated for how well they predicted life span based on the data from the metabolites in the blood plasma and how closely they aligned with health and aging markers. The scientists playfully named the metabolomic age derived from metabolite biomarkers “MileAge.” The MileAge delta measures the gap between a person’s MileAge and chronological age. If that gap is high, the person has accelerated aging.
“This metabolomic aging clock (MileAge) can be applied in research and may find use in health assessments, risk stratification, and proactive health tracking,” wrote the researchers.
There was a high-performance consistency among the top-performing AI algorithms, which included select tree-based ensembles and support vector regression. The top-performing algorithm used Cubist rule-based regression to calculate the MileAge delta to have results that were the most closely associated with markers for aging and health.
Similar to how older cars tend to have high mileage, people with accelerated aging had high MileAge deltas.
“Across most models, individuals with an older metabolite-predicted than chronological age, indicating accelerated aging, were frailer, had shorter telomeres, were more likely to have a chronic illness, rated their health worse, and had a higher mortality risk,” the researchers reported.
Interestingly, despite the finding that accelerated metabolomic aging had a clear link to higher mortality risk and poor health, the results showed that decelerated aging was not a reliable indicator for better health outcomes. The scientists caution that currently, certain metabolomics-based risk should mostly be used to spot patients with high risk.
“Aging clocks hold substantial promise for research on life span and health span extension, as they provide an aging biomarker that is potentially modifiable,” the scientists concluded.
With the proof-of-concept for MileAge at a systems level, the researchers suggest creating aging clocks based on tissues and cells as potential research avenues to traverse in the future.
Copyright © 2024 Cami Rosso All rights reserved.
