METRON: Metabolic Dynamic Perception Kolmogorov-Arnold Network for Biological Age Estimation

Area of Science:

• Biochemistry
• Computational Biology
• Gerontology

Background:

• Biological age, a physiological measure, is crucial for assessing health risks and aging.
• Steroid metabolomics provides insights into aging but faces challenges due to complex metabolic interactions.
• Existing models struggle to capture nonlinearities in metabolic networks.

Purpose of the Study:

• To introduce METRON, a deep learning framework for predicting biological age from steroid metabolomics data.
• To develop a model capable of capturing complex metabolite interactions and dependencies.
• To enhance the interpretability of biological age prediction models.

Main Methods:

• Developed METRON, a deep learning framework utilizing a Metabolite Interaction Perception Module (MIPM).
• Integrated a Group-Rational Kolmogorov-Arnold Network to capture intricate metabolic dependencies.
• Validated METRON's performance against existing machine learning and deep learning approaches.

Main Results:

• METRON demonstrated promising performance in predicting biological age compared to other methods.
• The framework identified Dehydroepiandrosterone (DHEA) as a known aging marker.
• 17-hydroxyprogesterone (17-OH-P4) was identified as a key signature linked to the hypothalamic-pituitary-adrenal axis.

Conclusions:

• METRON effectively predicts biological age using steroid metabolomics.
• The framework provides interpretability, uncovering key metabolic drivers of aging.
• METRON advances the understanding of metabolic contributions to the aging process.