Prediction of Multiple Degenerative Diseases Based on DNA Methylation in a Co-Physiology Mechanisms Perspective

• Published in: International journal of molecular sciences Vol. 25; no. 17; p. 9514
• Main Authors: Zhang, Li, Cai, Ruirui, Wang, Chencai, Liu, Jialong, Kuang, Zhejun, Wang, Han
• Format: Journal Article
• Language: English
• Published: Switzerland MDPI AG 01.09.2024
• Subjects: Accuracy; Age; Aging; Aging - genetics; Algorithms; Biomarkers; Brain cancer; Breast cancer; CpG Islands; Datasets; Deep learning; Disease; DNA Methylation; Epigenesis, Genetic; Epigenetics; Genomes; Humans; Kidney cancer; Liver cancer; Lung cancer; Machine learning; Neural networks; Neural Networks, Computer; Neurodegenerative Diseases - diagnosis; Neurodegenerative Diseases - genetics; Osteoarthritis; Osteoporosis; DNA methylation; deep learning; degenerative diseases; epigenetic clock
• ISSN: 1422-0067; 1661-6596; 1422-0067
• DOI: 10.3390/ijms25179514

Degenerative diseases oftentimes occur within the continuous process of aging, and the corresponding clinical manifestations may be neurodegeneration, neoplastic diseases, or various human complex diseases. DNA methylation provides the opportunity to explore aging and degenerative diseases as epigenetic traits. It has already been applied to age prediction and disease diagnosis. It has been shown that various degenerative diseases share co-physiology mechanisms with each other, clues of which may be gained from studying the aging process. Here, we endeavor to predict the risk of degenerative diseases in an aging-relevant comorbid mechanism perspective. Firstly, an epigenetic clock method was implemented based on a multi-scale convolutional neural network, and a Shapley feature attribution analysis was applied to discover the aging-related CpG sites. Then, these sites were further screened to a smaller subset composed of 196 sites by using biomics analysis according to their biological functions and mechanisms. Finally, we constructed a multilayer perceptron (MLP)-based degenerative disease risk prediction model, Mlp-DDR, which was well trained and tested to accurately classify nine degenerative diseases. Recent studies also suggest that DNA methylation plays a significant role in conditions like osteoporosis and osteoarthritis, broadening the potential applications of our model. This approach significantly advances the ability to understand degenerative diseases and represents a substantial shift from traditional diagnostic methods. Despite the promising results, limitations regarding model complexity and dataset diversity suggest directions for future research, including the development of tissue-specific epigenetic clocks and the inclusion of a wider range of diseases.