Detecting significant features in modeling microRNA-­target interactions

MicroRNAs (miRNAs) are small non-­coding RNA molecules mediating the translational repression and degradation of target mRNAs in the cell [1]. Mature miRNAs are used as a template by the RNA-­induced silencing complex (RISC) to recognize the complementary mRNAs to be regulated. Up to 60% of human genes are putative targets of one or more miRNAs. Several prediction tools are available to suggest putative miRNA targets, however, only a small part of the interaction pairs has been validated by experimental approaches. In addition, none of these tools does take into account the network structure of miRNA-­mRNA interactions, which involves collaboration and competition [2] effects that are crucial to efficiently predict the miRNA regulation effects in a specific cellular context. A first solution to consider collaboration effects is given by the web tool ComiR [3], which predicts the targets of a weighted set of miRNAs, provided the miRNA expression profile of the samples/tissues of interest. The analysis of the expression profile of the RNA fraction immunoprecipitated (IP) with the RISC proteins is an established method to detect which genes are actually regulated by the RISC machinery. In fact, genes that result over-­ expressed in the IP sample with respect to the whole cell lysate RNA, are considered as involved in the RISC complex, then miRNA targets. Here, we aim to find the features useful to predict which genes are overexpressed in IP, i.e. miRNA targets, without actually performing the IP experiments. To this purpose, we compiled and analyzed a novel high throughput data set suitable to unravel the features involved in the miRNA regulatory activities.