Genome-wide DNA Methylation Prediction Based On Neural Network

As important epigenetic phenomenon of human genome,DNA methylation is an important epigenetic regulation of gene expression in cell division and proliferation,development and aging of life,development of genetic diseases related to uniparental disomy,and carcinogenesis.Detection of DNA methylation status is an adjunct to the differential diagnosis of tumor types and can be used for the detection of tumor suppressor gene specific inactivation.A new generation of sequencing technology can achieve single-base single base resolution methylation detection,but the sequencing method is so expensive.Using computational methods to predict the status of DNA methylation is a hot spot of research on bioinformatics.In the theses,three news neural network models are designed to predict the status of DNA methylation and have higher accuracy.The main work is as follows:(1)Predicting DNA methylation status using artificial neural network model.According to the selected DNA methylation features by artificially extract to train artificial neural network model and predict the DNA methylation.(2)Predicting DNA methylation status using deep neural network and convolution neural network.A 5-layer depth neural network and a 4-layer convolution neural network are trained for DNA methylation prediction.All of features information of DNA methylation is input into the two depth learning models to complete the DNA methylation status.(3)Features are extracted by deep neural network and convolution neural network and then input to random forest classifier to predict the DNA methylation status.The DNA methylation features of depth neural network are extracted and input to the random forest,the extracted features are combined with the original features to apply to random forest.The DNA methylation features of convolution neural network are extracted and input to random forest,the extracted features are combined with the original features to apply to random forest.