On the Completeness of Existing RNA Fragment Structures.

The success of protein structure prediction by the deep learning method AlphaFold 2 naturally raises the question of whether similar success can be achieved for RNA structure prediction. One reason for the success in protein structure prediction is that the structural space of proteins, from the fragment level to the domain level, has been nearly complete for many years. Here, we examined the completeness of RNA fragment structural space at the di-, tri-, tetra-, and penta-nucleotide levels. We show that the number of non-redundant structural fragments at the tetra- and penta-nucleotide levels is in the midst of exponential increase, suggesting that the structural space currently observed in RNA is far from complete. Thus, more concerted efforts are clearly needed to improve the speed and methods of experimental determination of RNA structures to go beyond the limited structural space observed in RNAs. Moreover, the reference frame composed of three sugar-ring atoms near the base side (O4', C1', and C2') exhibits the least structural diversity among existing RNA structures, suggesting it as the most stable platform for building other parts of RNA structures.