A novel computational approach for genome-wide prediction of small RNAs in bacteria

Small regulatory RNAs (sRNAs) are the most abundant post-transcriptional regulators in bacteria. They serve ubiquitous roles that control nearly every aspects of bacterial physiology. Identification of important features from sRNAs sequences will guide the computational prediction of new sRNA sequences for a better understanding of the pervasive sRNA-mediated regulation in bacteria. In this study, we have performed systematic analyses of many sequence and structural features that are possibly related to sRNA properties and identified a subset of significant features that effectively discriminate sRNAs sequences from random sequences. we then used a neural network model that integrated these subfeatures on unlabeled testing datasets, and it had achieved a 92.2% recall and 89.8% specificity. Finally, we applied this prediction model for genome-wide identification of sRNAs-encoded genes using a sliding-window approach. We recovered multiple known sRNAs and hundreds of predicted new sRNAs. These candidate novel sRNAs deserve extensive study to better understand the sRNA-mediated regulatory network in bacteria.