Accessible Mirna Target Gene Prediction Algorithm Based On Markov Model

miRNA is a kind of important gene expression regulatory factors. The accurateprediction of target genes for miRNA is significant to study the function and themechanism of miRNA. Currently it is found a large number of miRNA, but the number ofinteracted target genes of which is small. Plant miRNA binding to mRNA with nearlyperfect complementary way, the target gene can easily be predicted. It is difficulty toprediction the animal miRNA target which only part of the base pairing. As themechanism of miRNA and the target is not clear, the algorithm based on the specificinteraction rules are not so satisfied. This paper studying the accession of mRNA andpropose a new miRNA target prediction algorithm.Firstly, according to the features that miRNA bind to mRNA with specific sequence.We design a probability model. A new miRNA target prediction algorithm Atar ispresented which based on the accessible of mRNA second structure. Atar allows and thereare at most a G:U base pair in the miRNA seed region. Construct second-order MarkovModel counting the numbers with a non-overlapping way that a given oligomer (specificnucleotide fragment) complementary to miRNA seed in the accession sites of 3’UTR.Rank the miRNA-3’UTR pairs according to the probability values that the certain numberoligomer appeared in 3’UTR and obtain the prediction results which ranked in the top N.Secondly, aimed at the high storage capacity and multi-dimensional disadvantage, weproposed a new miRNA-mRNA accession interaction site probability algorithm based onvariable length markov chain model. The algorithm alignments miRNA and mRNAaccession sites duplex and result a new sequence. The new sequence using differentcharacters represent different types of base pairs. Modeling the new sequence withvariable length Markov chain and improve the model by add smoothing techniques toprediction suffix trees. Judgment the category by compared the values of likelihood rateon the positive and negative class. The interaction site information then can obtained.Lastly, assessment the algorithm and other prediction algorithms in general criteria byusing the experimental verification target gene data in miRbase. The experiments resultsconfirmed that the proposed algorithm has high sensitivity and accuracy compared with other algorithms, it outperforms other algorithm.