Mirna Structure And Cancer Related Research Based On Computational Methods

With development in sequencing technology, there are a huge number of RNA data. However, the fact that RNA is not as stable as DNA has brought some difficulties in RNA experimental research. Using computational methods in analyzing the characteristics of various types of RNA is now one of the most important research fields in bioinformatics. MiRNAs are a class of single-stranded noncoding RNAs widespreading in flora and fauna, regulating the target genes expression by binding to specific region of their mRNA3’UTRs. MiRNAs are important non-coding RNA trans-regulatory factor, which are involved in various physiological processes, including embryonic development&developmental timing, cell proliferation&differentiation, apoptosis and hormones. Therefore, the variation of miRNA or misuse of miRNA regulation can lead to a variety of diseases, including cancer.Currently, computational methods applicated in miRNA research focused on two aspects, which are miRNA gene detection and miRNA target gene identification. The computational methods in miRNA structure research are lacking experimental data. The computational methods in miRNA function research are ineffective, due to the high false positive rate of miRNA target predictions, miRNA and diseases association analysis studies are not accurate. To solve the above problems, the following work was carried out:We developed a pre-miRNA stem-loop classifier based on support vector machine with radial basis kernel, using stem-loop structure characters. The classifier can effectively classify pre-miRNA stem-loops and non-miRNA precursor stem-loops. The Matthews correlation coefficient thereof achieved0.882and AUC of ROC curve was0.964.On the basis of the pre-miRNA stem-loop structure classifier, we further developed a classifier of Drosha enzyme site using support vector machine. The classifier also based on radial basis kernel, adopting the feature set integrated chemical kinetic. According to the testing result, the classifier of Drosha enzyme site can effectively predict Drosha enzyme cleavage sites on miRNA precursor stem-loops. Due to the integration of chemical kinetics, the classifier can more accurately distinguish real Drosha enzyme cleavage sites and adjacent sites. A two-layer svm classifier for Drosha processing sites was introduced by combining pre-miRNA stem-loop structure classifier and the classifier of Drosha enzyme site. The Matthew correlation coefficient reached0.9.We did a text mining on literature in MEDLINE and extract1,018miRNA-cancer associations between226miRNA families and20common human cancers from986avidence paper. Then we constructed a database named miCancerna which can be free assessed. Test result show that the text mining is stable and reliable. Comparing with existing databases and query technology, miCancerna provide more comprehensive miRNA-cancer association information. Besides, miCancerna also provided significant level for each miRNA-cancer association and presented significant miRNA-cancer associations in a bipartitel network which is helpful for the analysis of miRNA and cancer relationship.We further applied random walk with restart algorithms on miRNA-cancer association data in miCancerna to predict cancer-related miRNA. The leave-one-out test showed AUC of ROC curve was0.798, better than the current disease-related miRNA prediction algorithms.300randomized trials showed the prediction result is reliable. At last5candidate cancer-related miRNAs are predicted for each type of cancer,71cancer-related miRNA thereof got confirmed by later experiments or literature.Overall, the work we carried out filled some research gaps or gained improvements in bioinformatics research on miRNA.