Research Of MiRNA Prediction Algorithm Based On Manifold Structure

MiRNAs, first discovered in1993during studying the development and change ofC.elegans, are endogenous and single-stranded non-coding RNAs of only22nucleotides.They are derived from long precursors that fold into hairpin structures, and regulate thegene espression by combing the corresponding miRNA target. Studies have shown thatmiRNAs play fundamentally important roles in gene regulation in animals and plants.Discovery and identification of novel miRNAs is a hot topic of RNA genomics.Firstly, to the situation, the current miRNA prediction algorithms all require a largenumber of known miRNAs, but the existing miRNAs proved are very limited, this paperpresent a miRNA prediction algorithm based on geodesic distance and hidden markovmodel. The manifold structures of miRNAs high-dimensional data sets are studied to usegeodesic distance characterize the intrinsic geometric relationships of data points. Therand m walk based on the belief propagation on the hidden markov model was constructedto calculate the relevancy values of unknown samples following the relevancy updatingrule, which are sorted in decreasing order. The top n samples are selected as retrievedmiRNAs.Secondly, in order to study the contribution of the characteristics of miRNAs highdimensional data sets, and avoid over fitting problem caused by the high dimension, thispaper present a miRNA declines dimension algorithm based on distinguishing varianceembedding. This algorithm analysis the miRNA dimension from the manifold aspects andremove the characteristics value of low contribution to the prediction performance.Finally, this article uses the two algorithms to perform experiment on miRBase dataset. The experimental results and performance evaluate standards shown that the algorithmbased on geodesic distance and hidden markov model is feasible and effective, and hashigh precision and recall. MiRNA declines dimension algorithm based on distinguishingvariance embedding effectively reduce the dimension of miRNAs high dimensional datasets and acquire the low-dimensional representation.