ESPE Yearbook of Paediatric Endocrinology

Sci Adv. 2025;11(11):eadt2624. doi: 10.1126/sciadv.adt2624

Brief Summary: This study presents a novel deep neural network (DNN) model to predict biological age (BA) by integrating multi-omics data, with a focus on steroidogenesis pathway activity. The authors analysed 22 serum steroids from 148 healthy individuals aged 20 to 73 years, to develop and train the model. The DNN achieved accuracy for BA and identified cortisol as a key driver of aging and that smoking accelerates biological aging in males, highlighting the influence of lifestyle, sex/gender and stress-related hormones on aging.

This study introduces a biologically interpretable, pathway-based DNN model that advances the precision of biological age prediction by integrating steroid hormone metabolism, a critical but underutilized dimension in aging research. It shows that modeling steroidogenesis pathway activity can enhance biological age prediction, potentially providing a sensitive marker for early detection of physiological aging changes or monitoring interventions. By shifting the focus from passive chronological age to active biochemical processes, this study lays the groundwork for integrating functional endocrine biomarkers into the investigation of the multifaceted nature of aging and disease management.

On a larger picture, the suggested model provides a novel framework for personalized aging assessment using metabolomics. It demonstrates the utility of integrating biological pathways into machine learning models for improved interpretability and accuracy. It also offers potential for clinical applications in aging-related disease risk stratification and intervention monitoring.

The model relied on comprehensive serum steroid profiling, which may limit its applicability in settings with restricted biomarker availability. It did not account for circadian rhythms, dynamic physiological changes, or longitudinal fluctuations in serum steroid levels. Additionally, the study used a relatively small and demographically narrow sample and considered only limited lifestyle factors (e.g., smoking).

Further studies are needed to validate these findings, particularly in relation to environmental and behavioral influences as well as sex-specific differences.