scAgeClock: a single-cell transcriptome-based human aging clock model using gated multi-head attention neural...

Area of Science:

• Biogerontology
• Computational Biology
• Genomics

Background:

• Aging clock models are vital for assessing biological age and disease risk.
• Existing models lack single-cell resolution and struggle with tissue heterogeneity.

Purpose of the Study:

• To develop scAgeClock, a novel single-cell aging clock model using a gated multi-head attention neural network.
• To enhance the accuracy of biological age prediction across diverse human cell types and tissues.

Main Methods:

• Trained scAgeClock on over 16 million single-cell transcriptome profiles from 40+ human tissues.
• Utilized a gated multi-head attention neural network architecture.
• Developed a novel aging deviation index to analyze age patterns.

Main Results:

• scAgeClock achieved improved age prediction accuracy over baseline methods.
• Nearly 50% of tissue-level cell types showed prediction errors under 10 years.
• Identified key aging genes in pathways like ribosome, translation, and immune response.
• The aging deviation index revealed age deceleration in potent cells and tumor cells, and acceleration in skin cells.

Conclusions:

• scAgeClock offers a powerful tool for single-cell age prediction, accounting for cellular heterogeneity.
• The model provides insights into aging mechanisms and disease associations (e.g., COVID-19).
• Public availability of scAgeClock will foster further research in aging biology and interventions.