MicroRNA Prediction Based On Manifold Learning

RNA is the biological molecule with genetic information. MicroRNAs areendogenous and single-stranded non-coding RNAs of only22nucleotides. MicroRNAsregulate the gene espression by combing the corresponding miRNA target. Studies haveshown that MicroRNAs play fundamentally important roles in gene regulation in animaland plant cells. Nowadays, how to predict novel MicroRNAs is a hot topic.Firstly, in order to investigate the contribution of features for MicroRNA predictionalogrithm. This paper proposed triplet representation to extract sequence-structure features.Local sequence features are important in pre-miRNAs. The structure of sequence could becharacterized solely by36features.Secondly, machine learning is widely used, but it needs to describe the MicroRNA andnon-MicroRNA correctly. This paper proposed a novel ranking algorithm based onmanifold learning. MicroRNAs and their relationship can be modeled by a weighted graph,and the weight of edge equantifies the relation. Manifold ranking algorithm ranks the datawith respect to the intrinsic manifold structure collectively revealed by the given data. Theoutput of the manifold ranking algorithm is an ordered set.Lastly, due to the limitations of the research for the sequences in the Euclid space, it ishard to describe the internal structure of data set correctly. In order to overcome thisshortcoming, this paper predicted MicroRNA use kernel neighbor algorithm on statisticalmanifold. Feature vectors are mapped to polynomial space manifold according to dirichletcompound multinomial model. Each candidate sequences are classified by kernel neighboralgorithm on statistical manifold, combine with the result of manifold ranking to predictenew MicroRNA s.