ResnetAge: A Resnet-Based DNA Methylation Age Prediction Method

Abstract

Aging is a significant contributing factor to degenerative diseases such as cancer. The extent of DNA methylation in human cells indicates the aging process and screening for age-related methylation sites can be used to construct epigenetic clocks. Thereby, it can be a new aging-detecting marker for clinical diagnosis and treatments. Predicting the biological age of human individuals is conducive to the study of physical aging problems. Although many researchers have developed epigenetic clock prediction methods based on traditional machine learning and even deep learning, higher prediction accuracy is still required to match the clinical applications. Here, we proposed an epigenetic clock prediction method based on a Resnet neuro networks model named ResnetAge. The model accepts 22,278 CpG sites as a sample input, supporting both the Illumina 27K and 450K identification frameworks. It was trained using 32 public datasets containing multiple tissues such as whole blood, saliva, and mouth. The Mean Absolute Error (MAE) of the training set is 1.29 years, and the Median Absolute Deviation (MAD) is 0.98 years. The Mean Absolute Error (MAE) of the validation set is 3.24 years, and the Median Absolute Deviation (MAD) is 2.3 years. Our method has higher accuracy in age prediction in comparison with other methylation-based age prediction methods.

Keywords: CpG sites; DNA methylation; age prediction; deep learning.

Figure 1

Organizational chart of the data sample. The figure divides the dataset into ten parts according to the organization, with whole blood data and buccal data accounting for a larger share.

Figure 2

Overall flow chart of the model experiment.

Figure 3

Model architecture. The model has 14 convolution layers and 5 residual blocks, each of which consists of two convolution layers. In the model, the beta value of DNA methylation is used as input, and feature extraction is performed through 14 convolution layers. Finally, the 4D data are transformed into one-dimensional data to output the predicted age values.

Figure 4

Model visualization metric results. The horizontal coordinates in the figure represent the actual age of the sample, and the vertical coordinates represent the model-predicted age. The left graph shows the prediction results of the model in the training data, and the right graph shows the prediction results of the model in the test data.

Figure 5

Prediction errors of different methods in six organizations. The four violin plots represent the methods of ResnetAge, Horvath, Hannum, and ZhangAge, respectively. The six colors in each figure represent the age prediction errors of six types of tissues: whole blood, serum, saliva, buccal, uterine cervix, and T cell. The age prediction error is calculated by subtracting the actual age from the predicted age. The width of the violin plot represents the data density of the prediction error. The wider part indicates that the data distribution is denser, and the narrower part indicates that the data distribution is sparse.