| Timetable |
| Download Conference Program |
| Download All Abstracts |
| Zoom Access |
| Corporate Program |
Poster Presentations
Day 3, June 24（Tue.）
Room P (Maesato East, Foyer, Ocean Wing)
- 3P-PM-39
A steroid pathway-based DNN model for Biological Age prediction via LC-MS/MS steroid profiling
(IPR, Osaka Univ.)
oZi Wang, Qiuyi Wang, Kenji Mizuguchi, Toshifumi Takao
Aging involves the progressive accumulation of cellular damage, leading to systemic decline and age-related diseases. Despite medical advances, accurately predicting Biological Age (BA) remains challenging due to the complexity of aging and limitations of current models. This study introduces a novel approach for BA prediction using a Deep Neural Network (DNN) based on steroidogenesis pathways. Steroids were quantified using an in-house LC-MS/MS method, with data stratified by sex and allocated for training or independent validation. To address physiological and experimental variability, tailored scaling techniques preserved the relative proportions of steroid concentrations and ensured reliable dataset alignment. The DNN model incorporates a custom loss function to account for the progressive heterogeneity of aging—a factor largely overlooked in previous models. Moreover, the architecture is designed to capture biochemical processes within key steroid pathways, substantially enhancing biological interpretability. By leveraging critical markers such as cortisol, the model underscores the roles of stress-related and sex-specific steroids in aging. The resulting framework offers a robust tool for BA prediction across diverse datasets, providing new insights and supporting targeted strategies in aging research and disease management