DeepMethyGene: a deep-learning model to predict gene expression using DNA methylations

Gene expression influences various physiological functions and plays a critical role in disease progression. One of the key mechanisms governing gene expression is DNA methylation, an epigenetic modification that serves as a crucial regulator of gene expression. In this study, we proposed a ResNet-based adaptive regression convolutional neural network model, DeepMethyGene, to predict gene expression using DNA methylation information. We compared the performance of DeepMethyGene with the state-of-the-art prediction model, geneEXPLORE. Results showed that our DeepMethyGene outperformed geneEXPLORE in predicting the expression of most genes. Furthermore, our research revealed that the number of methylation sites and the average distance between these sites and the gene transcription start sites (TSS) significantly affect prediction accuracy. By gaining a deeper understanding of the complex relationship between methylation and gene expression, we can better predict disease progression and provide a theoretical basis for clinical interventions. The data and code of DeepMethyGene are available at https://github.com/yaoyao-11/DeepMethyGene.