RNA Structure Prediction: Advancing Both Secondary and Tertiary Structure Prediction.

摘要

RNAs can function without being translated into proteins. These RNAs adopt a structure or structures to perform these functions, and accurate prediction of structure is a valuable tool for understanding these functions. RNA structure is hierarchical, beginning with the primary sequence, then the secondary structure, i.e. the set of canonical pairs, and ultimately the tertiary structure, i.e. the three-dimensional structure.;One significant tool for prediction of secondary structure is the nearest neighbor model. This assumes the free energy change of forming a base pair depends on the identities of the pair and the adjacent pairs. Parameters were previously derived from optical melting on RNA duplexes where it was assumed all strands would be completely duplex or single-stranded. When individual base pairs are allowed to break as a function of temperature, the model does not agree with experiment. A new treatment of the data is presented. The probabilities of individual base pairs are calculated using a partition function, allowing internal loops and frayed ends. The parameters of the nearest neighbor model are recalculated using a nonlinear fit to the original data. These new parameters better fit the data and should provide improved structure prediction.;Homologous RNAs adopt similar structures. One important structural motif is the pseudoknot, a structure difficult to predict and often found near functionally important regions. Combining information from thermodynamics and homology, the TurboKnot algorithm presented here finds ∼80% of known base pairs, and ∼75% of predicted pairs were found in the known structures. Pseudoknots are found with half or better of the false-positive rate of other methods.;Finally, a novel protocol for RNA tertiary structure prediction employing restrained molecular mechanics and simulated annealing is presented. The restraints are from secondary structure, co-variation analysis, coaxial stacking predictions, and, when available, cross-linking data. Results are demonstrated on five different RNAs. The predicted structure is selected from a pool of decoy structures by maximizing radius of gyration and base-base contacts. This approach is sufficient to accurately predict the structure of RNAs compared to current crystal structures, as evaluated by root mean square deviation and the accuracy of base-base contacts.