Prediction and Evaluation of miRNA -- Target Gene Pairs Using K-means Clustering and Bipartite Graphs with Statistical Scoring

Identifying microRNAs (miRNAs) and their target genes plays an increasingly important role in better understanding the regulatory activities in the cell. Most computational methods focus on the sequence complementarity between miRNAs and target genes without using actual expression data, which, even when used, has been primarily just for validation of the predicted relationship between specific miRNAs and genes. Recent findings have shown that many targets are missed by sequence-based approaches. In this work, we present a robust method to predict and evaluate miRNA-gene pairs based on their positional (time-course) expression data from next-generation sequencing and DNA microarray. The method first uses K-means clustering to group miRNAs and genes respectively, and then assigns miRNA-gene pairs to a bipartite graph with statistical scoring. The method is tested by ten-fold cross validation on two datasets in Arabidopsis, achieving a performance of about 0.70 ROC score.