Modeling transcriptomic age using knowledge-primed artificial neural networks.

Area of Science:

• Biotechnology
• Computational Biology
• Genomics

Background:

• Machine learning age clocks predict biological age but lack transparency.
• Current models offer limited insight into the molecular mechanisms of aging.

Purpose of the Study:

• To develop a novel, interpretable age clock that couples predictive accuracy with biological insight.
• To create an artificial neural network model that reveals aging pathways from gene expression data.

Main Methods:

• Designed an artificial neural network incorporating prior biological pathway knowledge.
• Trained the model on gene expression data from skin tissue.
• Validated the model's ability to predict age and identify aging pathways.

Main Results:

• The interpretable age clock achieved high accuracy in predicting biological age.
• The model successfully identified and revealed the aging states of key biological pathways.
• The clock elucidated mechanisms of accelerated aging (Hutchinson-Gilford progeria syndrome) and longevity interventions (caloric restriction).

Conclusions:

• This novel age clock provides both accurate age prediction and mechanistic insights into aging.
• The interpretable model advances aging research by linking gene expression to biological processes.
• The approach facilitates the study of aging conditions and interventions through a transparent, knowledge-driven framework.