ResnetAge: A Resnet-Based DNA Methylation Age Prediction Method.

Area of Science:

• Epigenetics
• Computational Biology
• Gerontology

Background:

• Aging is a primary risk factor for many diseases, including cancer.
• DNA methylation patterns serve as indicators of cellular aging and can be used to develop epigenetic clocks.
• Accurate biological age prediction is crucial for understanding aging and developing clinical interventions.

Purpose of the Study:

• To propose a novel epigenetic clock prediction method, ResnetAge, with enhanced accuracy for clinical applications.
• To leverage deep learning, specifically a ResNet neural network, for predicting biological age.
• To validate the method's performance across diverse human tissues.

Main Methods:

• Developed ResnetAge, a deep learning model utilizing a ResNet architecture.
• Input data comprised 22,278 CpG sites, compatible with Illumina 27K and 450K arrays.
• Trained the model on 32 public datasets encompassing various tissues (e.g., whole blood, saliva, mouth).

Main Results:

• Achieved a Mean Absolute Error (MAE) of 1.29 years and Median Absolute Deviation (MAD) of 0.98 years on the training set.
• Reported an MAE of 3.24 years and MAD of 2.3 years on the validation set.
• Demonstrated superior age prediction accuracy compared to existing methylation-based methods.

Conclusions:

• ResnetAge offers a highly accurate and robust method for predicting biological age.
• Epigenetic clocks based on DNA methylation, like ResnetAge, hold significant potential as clinical biomarkers for aging.
• The findings support the utility of deep learning in advancing epigenetic age prediction for research and diagnostics.