Skip to Main Content
For rapid ribonucleic acid (RNA) tertiary structure prediction, innovative methods have been proposed that exploit hydroxyl radical cleavage agents in a high-throughput manner. In such techniques, it is critical to determine accurately which residue a specific cleavage agent interacts with, since this information directly reveals the residue-residue interaction points needed for structure inference. Due to lack of effective automated methods, the process of locating contact points has been mostly done manually, becoming a bottleneck of the whole procedure. To address this problem, we propose a novel computational method to determine residue-residue interaction points from 2-D electrophoresis profiles. This method combines the deconvolution method for signal detection and statistical learning techniques for filtering noise, thus boosting specificity and sensitivity in harmony. According to our experiments with over 2000 actual gel profiles, the proposed technique exhibited 56.44%-90.50% higher performance than traditional methods in terms of the accuracy of reproducing manual contact maps measured by the F-measure, a widely used evaluation metric. We expect that adopting the proposed technique will significantly accelerate RNA tertiary structure inference, allowing researchers to explore more structures in given time.