PRIME-BSPre: A genome-wide protein-RNA binding sites prediction method based on templates

Abstract

Identification of RNA binding sites that potentially interact with RNA-binding proteins facilitates a comprehensive analysis of protein-RNA interactions and enables further investigation into the mechanisms underlying RNA splicing and modification. However, the current experimental data remains limited in comparison to the vast family of RBPs, and deep learning prediction methods are inadequate for those RBPs lacking sufficient interaction data for training. Therefore, we present PRIME-BSPre, a genome-wide method for predicting protein-RNA binding sites based on templates that incorporate both RNA sequence and secondary structure as well as the tertiary structure of corresponding RBPs. We have successfully benchmarked our method on the human genome, demonstrating excellent prediction performance on RBP datasets beyond our library and robustness across cell lines. Additionally, we are pioneers in introducing the low Shannon entropy algorithm to describe binding preferences of RNA motifs. Our predicted results further support the hypothesis that RBPs preferentially bind RNA motifs with low complexity.