RNA Contact Prediction Based On Restricted Boltzmann Machine

RNA is one of the basic biological molecules in cells.Recent research found that RNA has catalytic and regulatory functions.In addition,RNA is closely related to the occurrence and development of diseases such as the nervous system,cardiovascular and cancer.The RNA function depends on its three-dimensional structure.Due to the RNA degradation and flexibility,it is difficult to determine the complex RNA tertiary structure by existing experimental techniques.Current RNA tertiary structure prediction methods are able to build small RNAs.However,it is difficult to predict large RNAs with complex topological feature.The nucleotide-nucleotide interaction information is able to improve the RNA tertiary structure prediction.The Direct Coupling Analysis(DCA)can infer nucleotide-nucleotide interactions from the RNA multiple homologous sequences.However,the prediction accuracy needs to be improved.Therefore,there is an urgent need to develop new methods to accurately predict RNA nucleotide-nucleotide interactions.In this thesis,we introduce a new method,DIRECT,to predict RNA interactions based on the Restricted Boltzmann Machine.DIRECT can utilize the RNA non-homologous structural characteristics and the RNA homologous sequence information to predict the nucleotide-nucleotide interaction.First,a Restricted Boltzmann Machine model is used to statistically analyze the non-homologous RNA structural characteristics.Then,the direct coupling analysis is used to identify the co-evolutionary information from the RNA homologous sequences.DIRECT combines the RNA co-evolutionary information with structural features to improve the nucleotide-nucleotide interaction prediction accuracy.We tested DIRECT on standard benchmarks.The results show that DIRECT improves the prediction accuracy with 41%and 18%compared with the traditional mfDCA and plmDCA models.Moreover,DIRECT significantly improves the long-range contacts with sequence distance more than 24nt.In summary,we developed one RNA interaction prediction method based on the Restricted Boltzmann Machine.DIRECT can significantly improve the RNA nucleotide-nucleotide interaction prediction accuracy,and further improve the RNA tertiary structure prediction.