RBPSpot: Learning on Appropriate Contextual Information for RBP Binding Sites Discovery

Abstract

Identifying RBP binding sites and mechanistic factors determining the interactions remain a big challenge. Besides the sparse binding motifs across the RNAs, it also requires a suitable sequence context for binding. The present work describes an approach to detect RBP binding sites while using an ultra-fast BWT/FM-indexing coupled inexact k-mer spectrum search for statistically significant seeds. The seed works as an anchor to evaluate the context and binding potential using flanking region information while leveraging from Deep Feed-forward Neural Network (DNN). Contextual features based on pentamers/dinucloetides which also capture shape and structure properties appeared critical. Contextual CG distribution pattern appeared important. The developed models also got support from MD-simulation studies and the implemented software, RBPSpot, scored consistently high for the considered performance metrics including average accuracy of [~]90% across a large number of validated datasets while maintaining consistency. It clearly outperformed some recently developed tools, including some with much complex deep-learning models, during a highly comprehensive bench-marking process involving three different data-sets and more than 50 RBPs. RBPSpot, has been made freely available, covering most of the human RBPs for which sufficient CLIP-seq data is available (131 RBPs). Besides identifying RBP binding spots across RNAs in human system, it can also be used to build new models by user provided data for any species and any RBP, making it a valuable resource in the area of regulatory system studies.