Misalignment of age clocks.

Area of Science:

• Biomedical Science
• Computational Biology
• Gerontology

Background:

• Biological aging is a complex, non-linear process with poorly understood biomarkers.
• Machine learning (ML) age clocks often assume linear progression, potentially obscuring natural biological patterns.
• Existing age clocks face challenges in accurately reflecting aging hallmarks like inflammaging.

Purpose of the Study:

• To clarify the trade-offs between mathematical optimization and biological interpretability in ML age clocks.
• To investigate the incoherence and biases in major DNA methylation (DNAm) age clocks.
• To explore the potential of non-linear ML models for a more biologically accurate aging trajectory.

Main Methods:

• Analysis of mathematical transformations in ML age clock construction.
• Quantification of misalignment between major DNAm clocks and actual DNAm changes.
• Development of an interactive visualization tool to illustrate clock errors.
• Evaluation of model coherence and bias toward specific cell fractions (e.g., leukocytes).

Main Results:

• Major conventional DNAm age clocks exhibit incoherence and are skewed toward leukocyte fractions.
• Rectifying model incoherence leads to a balanced model less skewed toward neutrophils.
• Improved models demonstrate enhanced detection of inflammaging.
• Significant misalignment exists between major DNAm clocks and actual DNAm changes.

Conclusions:

• Conventional linear ML age clocks may oversimplify biological aging, leading to inaccuracies.
• Addressing incoherence in DNAm clocks improves their biological relevance and detection of aging hallmarks.
• Non-linear ML approaches offer advantages for capturing the natural trajectory of aging directly from data.