Modeling And Prediction Of Pre-mRNA Alternative Splicing

Alternative splicing is known as one of the most important regulation processes in post transcription of eukaryotes. Although the basic patterns, processes of alternative splicing and the functions of some regulatory factors have been detected by biologists, the real regulatory mechanism of alternative splicing hasn't been revealed so far. Along with scientists' profound study on various kinds of diseases, especially cancers, as one of the major sources of diseases, alternative splicing of pre-mRNA is being emphasized by more and more scientists. Recently, with the development of High-throughput exon array and High-throuthput sequencing technologies, Genome-wide research on regulatory mechanisms of alternative splicing is becoming one of the frontiers and hottest topics. The contribution of this thesis is that it is the first model-based method to predict the regulatory factors and their functions during the process of alternative splicing.After the introduction of the basic molecular mechanism of alternative splicing, the latest technologies and methods studying alternative splicing are reviewed.Affymetrix exon array, as well as the statistical models and data processing procedures leading to the selection of tissue-specific alternative spliced exons are discussed in detail.Then, an linear alternative splicing model is designed, which considers the splicing variation between two different conditions as the combinatorial functions of multiple regulatory factors. The exon inclusion contribution scores are used to evaluate the controbution of different candidate motifs bound by splicing factors separately. Using ant colony optimization algorithm, the efficiency of motif searching can be highly improved, compared with exhaustive searching method.The linear model is capable of selecting the most important motifs from the motif candidates, very efficiently. This kind of model is very helpful for us to extracting the main information from the huge set of data from biological experiments. Based on the results of linear model, a model for the prediction of splicing factors' functions is designed based on fuzzy theory and supporting vector machine. After training using samples from previous data processing, this model can predict the functions of splicing factors from the following aspects:(a) Predict the regulatory function of a single splicing factor;(b) Evaluate the affection of this factor by other factors;(c) Predict the combinatorial functions of multiple regulatory factors.Since the complexity of alternative splicing, some ideas and methods to improve the performance and accuracy of model prediction are discussed in the end of this thesis.