An artificial intelligence (AI) model developed by researchers at Osaka University has resulted in a new method to determine biological age by using a simple blood test to analyze hormone (steroid) metabolism pathways. Details of the new method, which are published in Science Advances, show how using a deep neural network (DNN) for aging analysis could lead to more accurate health assessments and personalized interventions based on an individual’s aging processes.

“Our bodies rely on hormones to maintain homeostasis, so we thought, why not use these as key indicators of aging?” said co-first author Qiuyi Wang, PhD, an assistant professor at the Institute for Protein Research, University of Osaka.

To test their hypothesis, the research team focused on steroids, which play a crucial role in metabolism, immune function, and stress response. The researchers’ model leverages a DNN to analyze steroid hormone pathways—the first AI model that examines interactions between different steroid molecules. Instead of simply measuring steroid levels, which are known to vary widely from person to person, the model uses steroid ratios to provide what the researchers say is a more personalized and accurate assessment of biological age.

The new method can use a sample as small as five drops of blood to assess 22 key hormone steroids and their interactions, unlike other previous models, which often rely on broad biomarkers such as DNA methylation.

Biological age is a measure of how the body has aged, which is distinct from chronological age, or the number of years since birth. Chronological age does not account for the biological and lifestyle factors that can influence how quickly a person ages. Knowing biological age is important because it can provide a more accurate picture of a person’s health risks and help guide personalized care.

For this research, the investigators gathered blood samples from 148 individuals aged 20 to 73. Of these, 98 samples were used to train the model, and 50 samples were used to validate it. After this analysis, a notable finding was the role played by the stress-related steroid hormone cortisol. The researchers found that when cortisol levels doubled, biological age increased by approximately 1.5 times.

“Stress is often discussed in general terms, but our findings provide concrete evidence that it has a measurable impact on biological aging,” said co-corresponding authorToshifumi Takao, PhD, a professor of biochemistry at the Institute for Protein Research.

The model was designed to account for the increasing heterogeneity of aging since as people get older, their biological age can diverge more from their chronological age. This model also takes into account sex-specific differences in steroid metabolism, which could play a crucial role in understanding how aging processes differ between men and women. The researchers noted that their model could potentially aid in early disease detection and personalized wellness programs tailored to slow aging.

While there are other methods for assessing biological age using blood, this model’s combination of steroid analysis and deep learning distinguishes it from existing methods. Some tests, such as those measuring DNA methylation or protein levels, are already used to estimate biological age, but these models often overlook the intricate hormonal networks that play a key role in aging.

“Our approach reduces the noise caused by individual steroid level differences and allows the model to focus on meaningful patterns,” noted co-first author Zi Wang, a doctoral candidate in computational biology at Osaka University.

According to the researchers, this makes the test not only more accurate but also potentially more affordable and widely applicable for clinical use.

The next steps for the researchers include expanding their dataset and incorporating additional biological markers to improve the model’s accuracy and robustness. They also plan to explore the use of alternative normalization strategies, such as leveraging total cholesterol as a reference, to facilitate adaptation to datasets with fewer available steroid measurements.