Unsupervised learning of aging principles from longitudinal data

Area of Science:

• Gerontology
• Computational Biology
• Biomedical Data Science

Background:

• Aging is the primary risk factor for diseases and mortality.
• The precise relationship between physiological aging and lifespan remains unclear.
• Understanding aging dynamics is crucial for developing interventions.

Purpose of the Study:

• To develop a novel method for quantifying the aging process using longitudinal physiological data.
• To establish a predictive model for remaining lifespan based on aging dynamics.
• To investigate the correlation between the proposed aging indicator and established hallmarks of aging.

Main Methods:

• Utilized analytical and machine learning tools, specifically a deep artificial neural network with auto-encoder and auto-regression (AR) components.
• Developed a dynamic frailty indicator (dFI) to model organismal state instability.
• Applied the model to longitudinal blood test data from the Mouse Phenome Database.

Main Results:

• The dFI demonstrated an exponential increase over time, accurately predicting remaining lifespan in mice.
• The model's predictions aligned with observed late-life mortality deceleration.
• dFI changes correlated with key aging hallmarks, including frailty index, inflammation markers, and senescent cell accumulation.

Conclusions:

• The dynamic frailty indicator (dFI) provides a robust, data-driven measure of the aging process.
• dFI is sensitive to both life-shortening and life-extending interventions, validating its biological relevance.
• This approach offers a new framework for understanding and potentially modulating aging.