A structure-based model for the prediction of protein–RNA binding affinity

• DOI: 10.1261/rna.071779.119
• Publication Date: 2019-08-08T20:15:11Z
• Authors: Chandran Nithin; Sunandan Mukherjee; Ranjit Prasad Bah...
• Keywords: Models, Molecular; 0301 basic medicine; Binding Sites; Protein Conformation; RNA-Binding Proteins; Reproducibility of Results; Models, Theoretical; Article; Structure-Activity Relationship; 03 medical and health sciences; RNA, Transfer; Mutation; Nucleic Acid Conformation; RNA; Algorithms; Protein Binding

Protein–RNA recognition is highly affinity-driven and regulates a wide array of cellular functions. In this study, we have curated binding affinity data set 40 protein–RNA complexes, for which at least one unbound partner available in the docking benchmark. The covers range eight orders magnitude as well four different structural classes. On average, find complexes with single-stranded RNA highest affinity, whereas duplex lowest. Nevertheless, free energy gain upon ribosomal proteins lowest tRNA an average −5.7 cal/mol/Å 2 entire set. We train regression models to predict from physicochemical parameters interfaces. best fit model maximum error provided three interface parameters: relative hydrophobicity, conformational change hydration pattern. This has been used predicting on test set, generated using mutated structures yeast aspartyl-tRNA synthetase, experimentally determined Δ G values mutations are available. predicted empirical correlate experimental observations. study should be useful further development prediction methods. Moreover, developed enhances our understanding basis provides platform engineer interfaces desired affinity.