Active Learning for microRNA Prediction

MicroRNA (miRNA) are short, non-coding RNA involved in cell regulation at post-transcriptional and translational levels. Wet-lab experimental validation of predicted miRNA is a resource-intensive procedure. Therefore, a variety of computational approaches have been developed to increase prediction accuracy and reduce validation costs. While these methods are highly effective, they require large labelled training data sets, which are often not available for many species. Simultaneously, emerging high-throughput wet-lab experimental procedures produce large unlabeled data sets of genomic sequence and RNA expression profiles. Existing methods are unable to leverage these unlabeled data. This paper explores the application of active learning to maximize the utility of both labelled and unlabeled training data in microRNA prediction for the first time. Results across six diverse species show that our active learning approach is able to greatly improve classification performance using a small number of labeled instances, outperforming state-of-the-art methods under equivalent training data constraints.