Finding Condition-specific Target Module of MicroRNA in Time Series Transcriptome Data - using Gaussian Process Model and Spherical Vector Clustering -

Abstract

MicroRNAs, widely conserved small non-coding RNAs in several species, are important key regulators which mediate post-transcriptional gene silencing. As it is known that microRNAs are involved in many important processes from cell differentiation to apoptosis in recent studies, microRNA target prediction has been studied in various ways. The most typical way for predicting microRNA targets is to use nucleotide sequence features, which does not take into account condition-specific differences in transcript expression in cells. Therefore, a number of tools using transcript expression profiles in specific biological context have been developed to overcome the weakness of the traditional methods based on sequence features. But there are few proposed tools for time-series transcriptome dataset that provides dynamic expression patterns of microRNAs and their target mRNAs which can improve accuracy of target prediction.

In this paper, a new pipeline is proposed that predicts microRNA targets by integrating sequence feature and time-series expression profiles in specific experimental condition. For two datasets with different experimental conditions and cell types, condition specific target modules were predicted with our new pipeline for differentially expressed microRNAs that were reported from original papers. The context specificity of target modules was measured with three (correlation-based, target gene-based, network-based) evaluation criteria. MirTime showed good performance in three criteria more consistently than other microRNA target prediction methods using expression profiles.