IDEAL-Age: an interpretable deep learning framework for single-cell resolution profiling of immunological aging

Area of Science:

• Immunology
• Computational Biology
• Genomics

Background:

• Immunosenescence, or immune system aging, compromises defense against infections and vaccine efficacy.
• Bulk transcriptomic analysis masks the cellular diversity crucial for understanding immunosenescence.
• Existing methods lack the resolution to capture cellular heterogeneity in immune aging.

Purpose of the Study:

• To introduce IDEAL-Age, an interpretable deep learning framework for analyzing single-cell PBMC transcriptomes.
• To provide a high-resolution computational tool for deciphering systemic immune aging.
• To identify cellular roles and physiological transitions associated with immune aging.

Main Methods:

• Developed IDEAL-Age, a deep learning framework operating on single-cell PBMC transcriptomes.
• Benchmarked IDEAL-Age against 35 existing methods across multiple independent cohorts.
• Utilized the framework's interpretability to analyze gene contribution trajectories and cellular roles.

Main Results:

• IDEAL-Age demonstrated superior predictive performance compared to 35 other methods.
• The framework identified linear and non-linear gene contribution trajectories, revealing distinct physiological transitions.
• Application to systemic lupus erythematosus showed accelerated immunological aging linked to interferon-associated monocyte shifts.

Conclusions:

• IDEAL-Age offers a high-resolution computational approach to understanding immune aging at the cellular level.
• The framework's interpretability provides insights into the complex cellular dynamics of immunosenescence.
• IDEAL-Age can identify disease-specific patterns of accelerated immune aging.