Research On Exon Skipping Events Based On Machine Learning Methods

Gene alternative splicing(AS)is a complex and diverse process that will remove intron sequences and reorganize.exon sequences to generate the mature mRNA.AS allows one gene to compile multiple RNAs,each of which controls the synthesis and function of some corresponding proteins,which leading to the fact that limited genes control the synthesis of nearly infinite proteins.And the nature of gene has created the diversity of organisms on the earth.But at the same time,some unusual combinations of gene AS are likely to induce various deadly diseases,which maybe bring great disaster to human survival Therefore,the study of gene AS is necessary.AS is generally divided into five basic types according to the habit and it is estimated that more than 40%of AS event in the human are the exon skipping(ES)event,which make ES event prediction has become a research hot spot in bio informatics.After years of analysis and research,a lot of predicting ES event methods have been proposed.Generally,they can be roughly divided into two categories:the traditional experiment method and the calculation method.Because traditional experiment methods are costly,labor-intensive,inherent biases and limited coverage,so computational prediction of ES event are more popular and more trusted.By studying the methods of predicting ES event,we find some limitations which are the incompleteness of RNA-Seq data and gene sequences information,which may lead to unpredictable risk when predicting ES events.In order to try to overcome these limitations,in this thesis,we propose a new methods to predict ES event,which is based on RNA-Seq data,gene sequence information and the Rotation Forests.In this method,we emphasize on the advantages of the two kinds of data and extract these features that can indicate ES event,then analyzing and predicting ES event.Firstly,we construct a RS features consisting of RNA-Seq features extracted from RNA-Seq data and the sequence features extracted from the gene sequence information.Then based on the RS features,combined with Rotation Forests,we propose a new method called RotaF-RSES for predicting the ES event.To validate the effective of RotaF-RSES,one dataset is adopted from two human tissues,and the result indicates that the RotaF-RSES method can overcome the limitations of the two kinds of data,improve the prediction result and provide useful help for the prediction of ES events.