On the completeness of existing RNA fragment structures

Abstract

Success in protein structure prediction by the deep learning method AlphaFold 2 naturally gives arise the question if we can do the same for RNA structure prediction. One reason for the success in protein structure prediction is that the structural space of proteins at the fragment level has been nearly complete for many years. Here, we examined the completeness of RNA fragment structural space at dimeric, trimeric, tetrameric, and pentameric levels. We showed that the RNA structural space is not even complete at the di-nucleotide level, whereas the exponential increase of new structural fragments is observed at tetrameric and pentameric levels. Moreover, the number of backbone fragments found in RNA (2510) is far smaller than the number of backbone fragments found in proteins (6652). This further suggests that a structural space currently observed in RNA is far from complete, considering that the RNA backbone (6 torsion angles) has more degrees of freedom than the protein backbone (3 torsion angles with one nearly fixed). In addition, we found that the three-atom representation (one backbone atom C4 and two sidechain atoms C1 and N1) has the lowest number of structural fragments, suggesting it as the most "stable" structural frame for building up the entire RNA structure.