Analysis Of The RNA Differential Methylation And Co-methylation Pattern

The emergence of the methylated RNA immunoprecipitation sequencing(Me RIP-seq)technology,begining with the rapid development of high-throughput sequencing technologies,makes it possible to detect RNA epigenetic modifications in a large scale,which allows transcriptome-wide profiling of RNA methylation.The discovery of RNA differential methylation and co-methylation patterns from Me RIP-seq high-throughput data helps to reveal potential functions of m RNA methylation in regulating gene expression and shearing,and effectively guide cancer intervention.In this thesis,the methods of RNA methylation site prediction,RNA differential methylation and co-methylation pattern analysis are studied.The main contributions are as follows:1.The existing detecting algorithms of RNA methylation sites usually use manual features to represent RNA sequence and use shallow classifiers to classify.In this thesis,a method of detecting RNA methylation sites based on convolution neural network(MethylCNN)is proposed.Methyl-CNN algorithm constructs multi-modal deep convolution neural network models,and inputs RNA sequences,chemical characteristics and secondary structure features to predict RNA methylation sites.Under 5 fold cross validation,the sensitivity,specificity,Mathews correlation coefficient and accuracy of Methyl-CNN algorithm are all higher than the existing algorithm,which indicates that Methyl-CNN algorithm can detect RNA methylation sites effectively.2.To solve the RNA methylation of small samples,a method of RNA differential methylation analysis(DRME)is proposed under the condition of small samples.DRME uses the exome Peak algorithm to detect the methylation,then 2 independent negative binomial distribution of reads count of methylated are modeled to solve the transcriptional regulation and biological replicate samples within group differences,and estimated variance based on two-dimensional local regression to detect differential methylation of RNA.In simulated and real experiments,DRME has better detection effect than Fisher's exact test,and the differential methylation loci are statistically significant.3.In view of the low expression of RNA methylation,a new method of RNA differential methylation analysis(QNB)for small sample sequencing data is proposed.Different from DRME,background estimation is only based on input control samples.QNB combines input samples and immunoprecipitation samples to estimate gene expression to improve detecting rate of low expression genes.Then,4 independent negative binomial distribution models are used to model the reads count in the methylation area,and the RNA differential methylation is detected,which is based on two-dimensional local regression to estimate variance.In simulated and real experiments,QNB has a higher detection effect than the other RNA differential methylation detection methods.4.The multi-group data sets under different experimental conditions(cell types,tissues or stimulation,etc.)are comprehensively analyzed by using the four different clustering methods of the K-means ? hierarchical clustering? non-negative matrix factorizaion and Bayesian factor regression model.A consistency evaluation method of clustering results between two clusters is proposed to find the co-methylated regions,and the approach of gene ontology enrichment analysis is used to confirm the existence of co-methylation patterns.