The predictive value of double-stranded RNA for A-to-I editing detection

Structured AbstractO_ST_ABSMotivationC_ST_ABSAdenosine-to-inosine (A-to-I) RNA editing, a crucial reaction for many processes that contribute to transcriptome plasticity, is both widely common across the transcriptome and difficult to predict due to a lack of distinctive genomic characteristics that can be obtained and analyzed computationally. An exception to this is the secondary structure of RNA molecules, which has been shown to have a major impact on the selectivity and specificity of the enzymes responsible for A-to-I editing. Yet, this information is rarely used for the task of editing site prediction. ResultsHere, we demonstrated the value of using base-pairing probabilities of RNA nucleotides to classify genomic sites as A-to-I RNA editing sites, using large-scale truth data which we compiled and make available for use in training future models. Our analysis suggests that the span of four bases from -2 (upstream) to +1 (downstream) of a putative editing site is most informative in this regard. A classifier trained on base-pairing probabilities alone performed with a positive predictive value (PPV) of 0.68, a negative predictive value (NPV) of 0.64, and an area under the receiver operating characteristic curve (AUC) of 0.71. By identifying structure-related features that are informative for detecting A-to-I RNA editing sites and quantifying their predictive value, this work advances our understanding of A-to-I editing determinants. AvailabilityAll source codes and data are available at https://github.com/Ally-s-Lab/P-BEP