Research And Implementation Of Prediction Method Of Plant MiRNA And Its Function

MicroRNA(miRNA) have been widely attended in recent years and are endogenous small non-coding RNA whose length is approximately 22nt(nucleotides),and they take part in a number of important life processes of plants and animals and post-transcriptional regulation by inhibiting the corresponding target gene translation or degrading them.Interdisciplinary study based on the life sciences, computer science and statistics produced biological information and it has played an important role in making the miRNA research development rapidly. In this paper, the prediction of the pre-miRNA and miRNA mature,functional similarity between miKNAs and the prediction of miRNA functions were studied,and some innovative results were obtained. These results include the following three aspects:An integrated prediction model was proposed which can be used to predict plant pre-miRNA and miRNA mature. Currently,as most of prediction models have been developed for animals and human,and can only predict pre-miRNA. In order to resolve those problems,in this paper, new 152 features was proposed,and new feature selection algorithm,called B-SVM-RFE, was also proposed by improving the classical SVM-RFE, and an integrated prediction model, namely mirPlantPreMat, was obtained based on SVM and the best feature set selected by B-SVM-RFE. The proposed model obtained the better performance by comparing with other prediction models and also obtained satisfactory results on 9 different plant species. The above experiment results exhibit that the proposed prediction model not only has efficient and reliable performance but also good generalization ability.As the existing methods have some problems such as most can not be suitable for plant,can not obtain quantifiable outcomes and the calculation method itself has some defects, etc,in this paper, a calculation method, called PPImiRFS, was proposed for the above problems.The method computes the functional similarity between miRNAs based on the weighted protein-protein interaction network and related graph algorithms, and the network weight was calculated based on the semantic similarity of Gene Ontology(GO). An improved breadth-first search algorithm also was proposed. The PPImiRFS is significantly outperforms other methods by comparing with other methods.A prediction method of miRNA function based on the research results of the functional similarity between miRNAs and the prediction of protein function was proposed. The method constructed miRNA functional similarity network based the functional similarity scores calculated by PPImiRFS and using threshold selection algorithm based clustering coefficient.Transductive Multilabel classification algorithm was applied to that network to predict the potential function of miRNAs. Experiments show that the method achieved very satisfactory results on multiple evaluation criterions.