A statistical torsion angle potential improves the quality of NMR-based RNA structures

A growing consensus among structural biologists suggests that the backbone of RNA is rotameric, i.e, only a finite number of discrete backbone conformers are physically allowed. Upon surveying structures from the Protein Data Bank we found that NMR-based RNA structures contain, in general, a larger fraction of conformational outliers than high-resolution X-ray counterparts. Here, we develop a new statistical torsion angle potential for RNA by applying adaptive kernel density estimation on a database of high-resolution X-ray structures. When tested on a set of six RNA systems, we find that the statistical potential, implemented during molecular dynamics calculations, reduces the fraction of backbone conformational outliers and improves the agreement with cross-validated residual dipolar couplings.