Prediction of microRNA targets in Caenorhabditis elegans using a self-organizing map

Motivation: MicroRNAs (miRNAs) are small non-coding RNAs that regulate transcriptional processes via binding to the target gene mRNA. In animals, this binding is imperfect, which makes the computational prediction of animal miRNA targets a challenging task. The accuracy of miRNA target prediction can be improved with the use of machine learning methods. Previous work has described methods using supervised learning, but they suffer from the lack of adequate training examples, a common problem in miRNA target identification, which often leads to deficient generalization ability. Results: In this work, we introduce mirSOM, a miRNA target prediction tool based on clustering of short 3'-untranslated region (3'-UTR) substrings with self-organizing map (SOM). As our method uses unsupervised learning and a large set of verified Caenorhabditis elegans 3'-UTRs, we did not need to resort to training using a known set of targets. Our method outperforms seven other methods in predicting the experimentally verified C. elegans true and false miRNA targets.