Improved predictive algorithm of RNA tertiary structure based on GNN

RNA-targeted drugs are expected to push the third wave of industrial revolution in the field of drug development. The premise of the development of RNA-targeted drugs is to determine the three-dimensional spatial structure of the target RNA, but due to the biological characteristics of RNA, it is difficult to determine its structure by experimental means. This paper introduces a graph neural network-based improved predictive algorithm of RNA tertiary structure, IPARTS, which uses the latest macromolecular sampling method FARFAR2 under the Rosetta framework to learn the interatomic forces including non-bonded connected atoms, and realize the spatial structure of RNA is predicted at the atomic level. Our method surpassed other prediction methods, and completed the accurate prediction and efficient modeling of the tertiary structure of RNA. The accuracy of the loss function reached 2.204, which also provided new ideas for the structure prediction of other biological macromolecules.