PiRNA Identify Research Questions Based On Naive Bayes

At present, piRNA identification is one of hot issues in the research ofbioinformatics. Piwi-interacting RNA (piRNA) is an important class of smallnon-coding RNA (ncRNA) molecules with mostly25~32nt in length. piRNAs formRNA-protein complexes through interactions with piwi proteins and in correlationwith the function of RNA silencing. Now the researches on ncRNA are mainly dividedinto two aspects. First, large-scale identification of ncRNA, mainly with the aid ofcomputer to extract feature information from existing ncRNA, then identify thefeature information with genome-wide. Second, functional analysis of ncRNA usedgenomics and experimental methods. In this thesis, we collected existingexperimentally validated piRNA of human, rat, mouse, fruit fly and other modelspecies to construct the training set, constructed several feature-based models basedon nucleotide composition and physical and chemical classification of nucleotides,respectively. By Na?ve Bayes classifier with5-fold cross validation, our overallprediction accuracy achieves82%, which is higher than the feature of k-merfrequency, etc. The results showed that the feature combination model of k-merfrequency with nucleotide classification could serve as a good alternative way topiRNA prediction. The contents of this thesis are as follows:In chapter1, we introduced the main content of bioinformatics and the mainwork of this paper;In chapter2, we summarized the background of ncRNA prediction problems andthe current research status;In chapter3, we introduced some feature extracting and machine learningmethods, including k-mer frequencies, standardized word frequencies under a K-2order Markov Model and different classifications of four nucleotides, etc;In chapter4, the main contribution of the work–prediction of piRNA by Na?veBayes classifier;In chapter5, concluding remarks and future work.