基于机器学习的miRNA靶基因预测-信号与信息处理专业论文.docxVIP

II II Abstract With the discovery of RNA interference (RNAi), the role of non-coding RNA (ncRNA) in gene expression and regulation has been concerned unprecedentedly, and microRNA (miRNA) has become a research focus because of the close relationship with many regulation function of organisms. Generally, miRNAs regulate the gene expression by means of the interaction with their target genes, so the study of miRNA targets has significant meaning for understanding the regulation mechanism of miRNAs. The price of existing experimental technology of identifying miRNA targets is very high, what’s more, it easily suffers the laboratorial equipments, environment and high-tech operators influence. Therefore, computational approaches are essential for miRNA targets finding in genomes. In these years, large-scale computational approaches have been developed, such as sequence-based approaches, machine learning approaches and so on. However, the sequence-based approaches are appropriate for such cases that the experimental data is insufficient or the mechanism of identifying miRNA targets isn’t known. With the number of miRNA targets accumulating at a rapid pace, it is very hard to excavate effective information from these giant data, and simulating complicated miRNA functional mechanism is also not applicable. Therefore, in existing approaches, the machine learning approaches are most popular. These approaches have obtained some certain efficacy, but there still existed some problems, such as incomplete training samples, deflection of features and so on. In addition, the process of optimizing features and use of classifier are relatively simple. Aiming at these problems, in this article, a new method of predicting miRNA targets based on machine learning is put forward: Firstly, the miRNA targets getting from wireworm, by the improved method of selecting samples which is to construct the training samples sets using the method of mirWIP, combin