DeepMAge: A Methylation Aging Clock Developed with Deep Learning.

Area of Science:

• Biogerontology
• Computational Biology
• Epigenetics

Background:

• DNA methylation aging clocks are crucial tools in biogerontology research.
• Machine learning, particularly deep learning, shows promise for predicting human age from molecular data.
• Previous deep learning applications achieved high accuracy with blood biochemistry, transcriptomics, and microbiomics data.

Purpose of the Study:

• To evaluate the performance of deep learning in DNA methylation-based age prediction.
• To compare a novel deep learning aging clock (DeepMAge) against the established 353 CpG clock.
• To assess the biological relevance of the DeepMAge clock in relation to health conditions.

Main Methods:

• Developed DeepMAge, a neural network regressor utilizing blood DNA methylation profiles.
• Trained DeepMAge on 4,930 samples from 17 studies.
• Validated DeepMAge on an independent set of 1,293 samples from 15 studies.

Main Results:

• DeepMAge achieved an absolute median error of 2.77 years on the independent verification set.
• The performance of DeepMAge was compared to the 2013 353 CpG clock.
• DeepMAge demonstrated biological relevance by assigning higher predicted ages to individuals with specific health conditions.

Conclusions:

• Deep learning models, such as DeepMAge, offer a powerful approach for developing accurate DNA methylation-based aging clocks.
• DeepMAge shows comparable or superior performance to existing methods and possesses biological relevance.
• This technology has potential applications in biogerontology and understanding the molecular underpinnings of age-related diseases.