Inferring Multidimensional Rates of Aging from Cross-Sectional Data

Area of Science:

• Computational Biology
• Biostatistics
• Genomics

Background:

• Modeling individual change over time is crucial across sciences.
• Cross-sectional data, observing individuals once, limits traditional time-series analysis.
• Human aging research requires methods for analyzing such limited temporal data.

Purpose of the Study:

• To develop an interpretable latent-variable model for inferring temporal dynamics from cross-sectional data.
• To address the challenge of reconstructing individual trajectories from single observations.
• To enable the study of human aging using readily available cross-sectional datasets.

Main Methods:

• Proposed a nonlinear latent-variable model where individual features evolve based on a low-dimensional, linearly-evolving latent state.
• Introduced an order-isomorphic constraint on the nonlinear function for model identifiability.
• Demonstrated identifiability from cross-sectional data under known time-independent variation distributions.

Main Results:

• The model successfully reconstructs observed data from the UK Biobank human health dataset.
• Learned interpretable aging rates associated with specific diseases.
• Identified aging rates linked to mortality and known aging risk factors.

Conclusions:

• The developed model effectively learns temporal dynamics from cross-sectional data, overcoming limitations of traditional methods.
• Provides a powerful tool for studying human aging and disease progression using observational health records.
• Offers interpretable insights into biological aging processes and their health-related consequences.