| microRNA is a kind of non-code RNA, which is single-stranded and has about 22-24 nucleotides. It is formed from pre-microRNA, and pre-microRNA is processed by endonuclease Dicer and Dicer-like-1 which have RNase III activity. microRNA controls gene expression by inducing mRNA's splice or suppressing its translation. microRNA not only controls about one third of human gene, but also cell's proliferation and differentiation,death,early development and metablolic process. Research shows that it has close relationship with cancer. The research of microRNA is helpful to realize the relationship of genes, what's more, it's helpful to research gene's function and biology's discovery. Although microRNA exists in 55 species, those founded yet are much less than that exist in real. There are a big amount of microRNAs needed to be founded.There are two main methods to predict microRNA:cDNA clone and computional prediction. The former one is mainly and it's directly and reliably, but it's hard to clone microRNAs those express in different period and in special organize and cell line. Computional prediction is not effected by expression period,expression level or organize specificity.This paper puts forward a prediction method based on machine learning, It is called ACO + SVM. For the long sequence and stem-loop structure of pre-microRNA,we extract features from its sequence and structure characteristics. This paper constructs a SVM classifier to distinguish positive and negative pseudo pre-microRNA. For the good feature of approximation and generalization, we choose SVM to train the classification. To improve the performance of the classifier, ant colony optimization was employed to search for C and g, which were two important parameters for SVM classifier. Experimental results indicate that the method ACO + SVM not only predicted human pre-microRNA efficiency, but also had higher accuracy in other species. Compared with other similar methods, it has better sensitivity and specificity. |
PDF Full Text Request