There are significant differences in the rate of aging between individuals. Some people show signs of aging before reaching middle age, with a protruding belly and fatigue. Others appear to freeze in time, looking much younger than their actual age in terms of appearance and physical condition.
Traditionally, the medical community often assesses "biological age" using biomarkers such as DNA methylation, telomere length, or proteomics. These methods provide insight into an individual's health by comparing biological age to chronological age, offering guidance for health interventions.
However, these methods often overlook a key system — the hormone network. A research team from Osaka University in Japan, in a study published in Science Advances, has developed a new approach using artificial intelligence (AI) to integrate steroid metabolism pathways, requiring just five drops of blood to accurately calculate biological age, providing a fresh perspective on aging differences.
The research team developed a deep neural network (DNN) model, which innovatively incorporates the interaction of 22 key steroid hormones into the analysis framework, making it the first AI model to explicitly explain the interactions between different steroid molecules. Steroid hormones were chosen because they more accurately reflect the aging process.
Steroid hormones (such as cortisol, sex hormones, and vitamin D derivatives) are core signaling molecules that regulate metabolism, immunity, and stress responses.
They directly regulate gene expression through nuclear receptors, affecting key aging-related pathways such as cell proliferation, inflammation, and energy distribution. Since steroid synthesis depends on the collaboration of multiple organs such as the adrenal glands, liver, and gonads, their metabolic profile can comprehensively reflect the body's physiological state.
In contrast, traditional aging studies focus on specific organs (such as the brain or cardiovascular system), failing to provide a comprehensive view of the body's overall condition.
Pathway-based DNN model for BA prediction from serum steroid profiling via LC-MS/MS
Unlike traditional methods that rely on absolute hormone levels, this model focuses on the changes in hormone ratios, quantifying the dynamic balance of the hormone network through blood samples.
The focus on ratio changes is due to the significant individual variation in hormone levels, which also fluctuate due to factors such as circadian rhythms. The ratio relationship between hormones is more stable. By analyzing these ratios, the model can effectively avoid the interference of individual baseline differences.
The study found multiple steroid biomarkers related to aging, with cortisol (COL) being particularly significant. Although these biomarkers have been mentioned in previous studies, this correlation study revealed more mysteries, showing that COL is positively correlated with biological age.
This finding fully supports the hypothesis that cortisol serves as a biomarker of stress, reflecting cumulative physiological damage.
Among lifestyle factors, smoking has a significant impact on accelerating the aging process in men. However, the validation cohort lacked data on behaviors such as alcohol consumption and diet. The results showed that only male smokers exhibited faster aging compared to non-smokers, suggesting that the higher frequency of smoking among men may be a contributing factor. In contrast, the aging effects in female smokers might be masked due to lower smoking frequency and unknown confounding factors. Future research with larger cohorts and more comprehensive lifestyle data will be needed to clarify the correlation.
Performance of the DNN model and smoking impact on BA prediction.
This study, through the deep integration of steroid metabolism networks and AI models, has pioneered a new paradigm for systematically analyzing the heterogeneity of aging. Its core value lies not only in the development of a highly accurate biological age prediction tool but also in uncovering the interactions of aging across three dimensions: stress, gender, and metabolism, providing theoretical foundations for personalized anti-aging interventions. Future research may require larger-scale, dynamic data verification to promote the model from the laboratory to clinical practice.
[1]. Qiuyi Wang et al. ,Biological age prediction using a DNN model based on pathways of steroidogenesis.Sci. Adv.11,eadt2624(2025).DOI:10.1126/sciadv.adt2624
 |
 |
 |
