Effective Clustering Of MicroRNA Sequences

microRNAs (miRNAs) are endogenous small non-coding RNAs that about19to24nucleotides in length. Since the first discovery of miRNA, lin-4, in1993, many studies have sought to identify miRNAs experimentally or computationally in var-ious species. Therefore, the number of miRNAs deposited in miRBase has been increasing approximately exponentially. miRNAs can act with the Argonaut fami-ly of proteins to regulate target messenger RNAs (mRNAs) post-transcriptionally. Hence, they play a key role in gene regulation and development of animals and plants. Currently, researches mainly focus on the function study of single miR-NA. Considering that members in the same miRNA family might participate in same pathway or regulate same targets and thus share similar biological functions, people can explore useful knowledge from high quality miRNA family architecture.In this article, we developed an unsupervised clustering method miRCluster to automatically group miRNAs. It is useful and efficient as no prior family architecture information is required and some dimension reduction methods are used. In order to evaluate this method, several data sets were constructed from miRBase. Results showed that miRCluster is effective in miRNA clustering:1. Efficiently arranging miRNAs. It can identify354of394families in miR-Base16with an accuracy of92.08%.2. Identifying new families. Compared to miRBase16, there are72new families and2383new miRNAs added to miRBase17. With miRCluster,9of all10newly-added families (contain no less than4members) are recognized.3. Effectively assigning unclassified miRNAs. By far,30%mature miRNAs registered in miRBase17are unclassified. With miRCluster, over85%un-classified miRNAs can be assigned to certain families, while44%of these miRNAs distributed in300novel families.