Ageing is an inevitable process that affects every individual differently. While chronological age is simply the number of years a person has lived, biological age reflects the body’s functional state. Two people of the same chronological age may have very different biological ages due to variations in lifestyle, genetics, and environmental influences. Understanding biological age can help predict health risks and ageing-related diseases more accurately than chronological age.
Traditional methods for estimating biological age relied on external indicators such as grip strength, lung function, and cognitive ability. While useful, these methods lacked precision and consistency. Over time, researchers have developed more sophisticated techniques that examine biomarkers at the molecular level. One promising approach involves analyzing steroid hormones, which play crucial roles in metabolism, stress response, and ageing.
The Role of Steroids in Ageing
Steroid hormones are biochemical messengers that regulate a wide range of bodily functions, including metabolism, immune response, and reproductive health. As people age, the levels of these hormones change, leading to shifts in physiological processes. Corticosteroids, which are involved in stress regulation, tend to increase with age, while sex hormones such as estrogen and testosterone often decline.
These hormonal changes can impact everything from muscle mass and bone density to cognitive function and cardiovascular health. Because of their strong connection to ageing processes, steroid hormones offer a valuable window into biological age. Scientists have long recognized the potential of steroid profiling, but previous models struggled to capture the complex interactions between these hormones and the ageing process.
A Deep Learning Approach to Biological Age Prediction
To improve the accuracy of biological age prediction, researchers developed a deep neural network (DNN) model that analyzes steroid hormone pathways. Deep learning algorithms can process vast amounts of biological data, uncovering hidden patterns that traditional statistical models might miss.
In this study, researchers analyzed 22 different steroid hormones from the serum samples of 148 individuals aged 20 to 73. They divided the data into training and validation groups, using 98 samples to train the DNN and 50 samples for independent validation. The model incorporated information about sex-specific variations in steroid metabolism, ensuring that the predictions accounted for differences between male and female ageing patterns.
Unlike traditional methods that focus on chronological age, this model considered the biological pathways that regulate ageing. By mapping the relationships between different steroids, the DNN provided a more nuanced and biologically meaningful prediction of age.
Capturing the Complexity of Aging
One of the most significant findings of this research was that ageing is not a uniform process. The model confirmed that the gap between chronological and biological age increases over time, meaning that individuals age at different rates. Some people experience accelerated ageing due to lifestyle factors, genetics, or environmental exposures, while others maintain a younger biological age despite advancing years.
This variability is particularly important for understanding age-related diseases. Many chronic conditions, such as Alzheimer’s, osteoporosis, and cardiovascular disease, develop due to physiological ageing rather than simply the passage of time. By identifying individuals with an advanced biological age, doctors could intervene earlier to reduce the risk of these conditions.
Impact of Smoking on Biological Ageing
The study also examined the relationship between smoking and biological age. Smoking has long been associated with premature ageing, but its effects can vary based on sex and individual differences in metabolism.
Interestingly, the researchers found that smoking accelerated biological ageing in males but had a less pronounced effect in females. This discrepancy could be due to differences in smoking habits, lifestyle factors, or hormonal influences. Men tend to smoke more frequently than women, which may contribute to the stronger effect observed in male participants.
While the study did not analyze other lifestyle factors in detail, it highlights the need for further research into how behaviors like diet, alcohol consumption, and physical activity influence biological age. Future studies could integrate additional biomarkers to provide a more comprehensive view of how lifestyle choices shape ageing.
Identifying Key Ageing Markers
One of the major contributions of this study was the identification of key hormonal markers associated with ageing. Among the 22 steroids analyzed, cortisol emerged as a significant predictor of biological age.
Cortisol is a stress hormone that plays a vital role in regulating metabolism, inflammation, and immune function. Elevated cortisol levels have been linked to chronic stress, increased fat accumulation, and cognitive decline—factors that contribute to ageing. The study found a strong positive correlation between cortisol levels and biological age, reinforcing its role as a key biomarker of ageing.
Other hormones, such as testosterone and estrogen, also influenced biological age. In males, testosterone levels were closely linked to ageing, while in females, estrogen-related markers played a more significant role. These findings highlight the importance of considering sex-specific differences when developing ageing models.
Refining the Accuracy of Ageing Models
To ensure the accuracy of their predictions, the researchers employed advanced data scaling techniques. They used a cumulative distribution function (CDF)-based proportional scaling approach to align the steroid concentration data across different samples. This method preserved the relative proportions of steroid levels while minimizing variability caused by experimental conditions.
One of the challenges in ageing research is that biological markers can be influenced by external factors such as circadian rhythms, dietary habits, and stress levels. By using a carefully designed scaling method, the researchers reduced these potential sources of bias, improving the reliability of the model.
Their approach also addressed one of the common criticisms of deep learning models—lack of interpretability. Many machine learning models function as "black boxes," making it difficult to understand why certain predictions are made. By structuring their DNN around known steroid metabolic pathways, the researchers ensured that the model’s outputs aligned with established biological knowledge.
Future Directions for Ageing Research
While this study represents a significant step forward in biological age prediction, there is still room for improvement. One limitation of the current model is that it focuses exclusively on steroid hormones. While steroids provide valuable insights into ageing, other biological markers—such as DNA methylation patterns, protein expression levels, and metabolic byproducts—could enhance predictive accuracy.
Longitudinal studies that track individuals over time would also help refine the model. Ageing is a dynamic process, and a single snapshot of steroid levels may not fully capture the long-term changes that occur. Future research should incorporate repeated measurements to observe how steroid profiles evolve with age.
Additionally, integrating lifestyle and environmental factors could improve predictions. While this study highlighted the impact of smoking, other factors such as diet, exercise, and stress management play crucial roles in ageing. A more holistic approach that combines molecular data with behavioral information could provide deeper insights into how people can influence their ageing trajectories.
Practical Applications of Biological Age Prediction
The ability to accurately estimate biological age has significant implications for medicine and public health. Doctors could use these models to assess an individual’s ageing rate and recommend personalized interventions. For example, someone with an accelerated biological age might benefit from targeted lifestyle changes, stress management strategies, or medical treatments to slow down ageing-related decline.
This approach could also aid in drug development and clinical research. By identifying individuals with different ageing rates, scientists could study how various interventions affect biological age. This could lead to the development of new therapies aimed at slowing ageing and preventing age-related diseases.
For the general public, understanding biological age could encourage healthier lifestyle choices. People who see a discrepancy between their chronological and biological age might be motivated to adopt healthier habits, such as improving their diet, increasing physical activity, or quitting smoking.
Conclusion
This study demonstrates the power of deep learning in biological age prediction. By analyzing steroid hormone pathways, researchers developed a model that captures the complexity of ageing and highlights key biomarkers such as cortisol. Their findings confirm that ageing is highly individualized, with factors like stress, smoking, and hormone levels influencing biological age.
As research in this field advances, these models could become essential tools for healthcare professionals, researchers, and individuals looking to understand and manage their ageing process. With continued improvements, biological age prediction could revolutionize the way we approach ageing, shifting the focus from treating age-related diseases to proactively managing the ageing process itself.
The study is published in the journal Science Advances. It was led by researchers from Osaka University.
