State-of-the-art methods for predicting novel trans RNA-RNA interactions use the so-called accessibility as key concept. It estimates whether a region in a given RNA sequence is accessible for forming trans interactions, using a thermodynamic model which quantifies its secondary structure features. RNA-RNA interactions are then predicted by finding the minimum free energy base pairing between the two transcripts, taking into account the accessibility as energy penalty. We investigated the underlying assumptions of this approach using the two methods RNAPLEX and INTARNA on two datasets, containing sRNA-mRNA and snoRNA-rRNA interactions, respectively. We find that (1) known trans RNA-RNA interactions frequently overlap regions containing RNA structure features, (2) the estimated accessibility reflects sRNA structures fairly well, but often disagrees with structures of longer transcripts, (3) the prediction performance of RNA-RNA interaction prediction methods is independent of the quality of the estimated accessibility profiles, and (4) one important overall effect of accessibility profiles is to prevent the thermodynamic model from predicting too long interactions. Based on our findings, we conclude that the accessibility concept to the minimum free energy approach to predicting novel RNA-RNA interactions has conceptual limitations and discuss potential ways of improving the field in the future.

Splicing is one key mechanism determining the state of any eukaryotic cell. Apart from linear splice variants, circular splice variants (circRNAs) can arise via non canonical splicing involving a back-splice junction (BSJ). Most existing methods only identify circRNAs via the corresponding BSJ, but do not aim to estimate their full sequence identity or to identify different, alternatively spliced circular isoforms arising from the same BSJ. We here present CYCLeR, the first computational method for identifying the full sequence identify of new and alternatively spliced circRNAs and their abundances while simultaneously co-estimating the abundances of known linear splicing isoforms. We show that CYCLeR significantly out-performs existing methods in terms of sensitivity, precision and quantification of transcripts. When analysing D. melanogaster data, CYCLeR uncovers biological patterns of circRNA expression that other methods fail to observe.