Ab initio Prediction of RNA Nucleotide Interactions with Backbone k-Tree Model.

Recently there have been many revelations of importance of RNAs to cellular functions and thus fast-growing interests in computational prediction of RNA tertiary, which is a significant challenge nevertheless. Even for a short RNA sequence, the space of tertiary conformations is immense; existing methods to identify native-like conformations mostly resort to random sampling of conformations for computation feasibility. To avoid compromise in accuracy, such methods may also be guided with knowledge of local tertiary motifs or secondary structure. However, due to lack of effective, global constraints on the RNA tertiary structure, existing methods have yet to deliver the desired prediction accuracy for RNAs of length beyond 50. In this work, we approach the ab initio tertiary structure prediction problem with a key step to predict nucleotide interactions that constitute the desired tertiary structure. Our work is established on a novel graph model, called k-tree, to constrain nucleotide interaction relationships in RNA tertiary structure. We show that, effectively guided by the k-tree model, a set of nucleotide interactions can be optimally and efficiently predicted from the query sequence. Our work also benefits from recent studies that revealed intrinsic and detailed roles of nucleotide interactions in the tertiary structure, including the fine-grained geometric nomenclatures and rich families of nucleotide interactions. Test results demonstrate that our method can predict with a high accuracy nucleotide interactions that constitute the tertiary structure of the query sequence, rendering a feasible solution toward ab initio prediction of tertiary structures.