Integration,Evaluation And Optimization For RNA Tertiary Structure Prediction Methods

Ribonucleic acid(RNA)is one of the most important biomolecules in living cells,which can regulate a variety of biological processes,including gene regulation,catalysis of biochemical reactions and communication between cells.These novel and rich functions of RNA are determined by their tertiary structure,so understanding their precise tertiary structure can better understand their function.Therefore,it is very important to study the tertiary structure of RNA to explain the biological function and regulation mechanism of RNA.Due to the limitations of the experimental methods,which are complicated,time-consuming,expensive and require special equipment,modeling by computational methods to predict the tertiary structure of RNA molecules has become more and more important.The results of RNAPuzzles competition showed that the method could be used to model the tertiary structure of RNA effectively,but there was still room for improvement.As more and more RNA tertiary structure prediction methods have been designed,it has become an important research topic to evaluate and optimize these prediction methods to more accurately predict RNA tertiary structure.The main research contents of this paper include:(1)In order to better integrate and analyze the existing RNA tertiary structure prediction methods,seven prediction methods with user-friendly interface(3dRNA,RNAComposer,MCFold/MC-Sym,Vfold,SimRNA,i Fold,FARFAR2)and the DSSR structure analysis program are integrated into a new RNA tertiary structure prediction tool.This integrated tool enables multi-method prediction of tertiary structure of RNA based on sequence and secondary structure,as well as comprehensive image display and structural analysis of the predicted structure.In addition,for molecules with experimental structures,the root mean square deviation between each predicted structure and experimental structure can be directly given,which is convenient for subsequent evaluation of existing methods.The integrated tool improves the time efficiency of structural prediction and analysis for practical applications such as drug design and gene therapy.(2)Based on the above RNA tertiary structure integration tool,we further constructed a dataset based on the RNA-Puzzles contest and evaluated the prevailing RNA tertiary structure prediction methods in combination with the existing RNA dataset containing 43 RNA structures of different length types.The prediction methods were compared from the aspects of RNA molecular size,molecular type and secondary structure type.The results showed that Vfold method predicted fewer RNA molecules,but its prediction effect was the best,and its dependence on molecular length was significantly lower than other prediction methods.The accuracy of i Fold and SimRNA methods was most dependent on RNA length.The accuracy of MC-Fold/MC-Sym structure prediction is relatively stable.RNAComposer method has good overall prediction accuracy,but the accuracy of prediction structure fluctuates greatly.(3)Through a comprehensive evaluation of existing RNA tertiary structure prediction methods,we will further design optimization strategies based on the strengths and weaknesses of each method to achieve more accurate prediction of RNA tertiary structure.Preliminary results show that for molecules with sequence length greater than 100 nt,the structure predicted by the 3dRNA method will dominate.For riboswitches,the accuracy of the SimRNA method is low.For RNA with simple structures such as hairpin ring and inner ring,the structures predicted by Vfold and RNAComposer methods could be preferred.However,for complex RNA(such as Pseudoknots and multi-branched rings),the structure predicted by SimRNA method is less reliable and can be omitted.We will select effective RNA structural features based on the above analysis results,and use machine learning or statistical learning methods to establish predictive method classification model.Work on this part is in progress.