Effective clustering of microRNA sequences by N-grams and feature weighting

MicroRNA (miRNA in short) is a kind of small RNAs that acts as an important post-transcriptional regulator with the Argonaute family of proteins to regulate target mRNAs in animals and plants etc. Since its first recognition as a distinct class of small RNA molecules in the early 1990s, tens of thousands of miRNAs have been identified experimentally or computationally. Currently, the focus of miRNAs study is on single-miRNA functions that usually result in gene silencing and repression. With the rapid increase of miRNAs, biologists have manually organized these miRNAs into biologically meaningful families to facilitate further study. As the members in the same family tend to share similar biochemical functions, a high quality family organization will shed lights on the functions of unknown miRNAs. However, manually grouping large amounts of miRNAs is not only time-consuming but also expensive. In this paper, we employ a clustering method with N-grams and feature weighting to automatically group miRNAs into separate clusters (families). Our method is evaluated with datasets constructed from the online miRNA database miRBase. Experimental results show that the clustering method can successfully distinguishes most miRNA families, and outperforms the traditional K-means clustering algorithm and the average-link clustering approach.