Semi- supervised ensemble learning to boost miRNA target predictions.

Abstract

The huge amount of data produced by the advent of Next Generation Sequencing (NGS) technologies is providing scientists with an unprecedented potential to investigate and shed light on remote secrets of genomes. We have developed a new tool based on biclustering techniques, i.e. HOCCLUS2 which is able to significantly correlate multiple miRNAs and their predicted targets to detect potential miRNA:mRNA regulatory modules. However, experiments performed on predicted interactions led to observe that the noise (i.e., false positives) introduced by prediction algorithms can substantially affect the significance of the discovered modules. In order to overcome this issue, we have developed a probabilistic method which is able to build a more reliable dataset, combining data produced by several well-known prediction algorithms. The main goal of this work is to combine the prediction score of several prediction algorithms in a single stronger classifier, in order to improve the reliability of the obtained predictions. This tool could greatly help in the interpretation of NGS miRNAs profile analysis with respect to their effects by using genome-wide predictions of their targets.