Abstract: Protein is a kind of substance with spatial structure.The main goal of protein structure prediction is to extract effective information from existing large-scale protein datasets,so as to predict the structure of proteins in nature.At present,one of the problems in protein structure prediction experiments is the lack of data sets that can further reflect the spatial structure of proteins.Although the current mainstream PDB(protein data bank) is experimentally measured,it does not utilize the spatial characteristics of proteins,and there are problems of doping nucleic acid data and partial data is incomplete.In view of the above pro-blems,this paper studies the prediction of protein from the perspective of spatial structure.Based on the original PDB,the Hohai graphic protein data bank is proposed.The dataset expresses the spatial structure characteristics of proteins based on the graph structure.Based on the traditional Transformer network model,relevant protein structure prediction experiments are carried out on the new dataset,and the prediction accuracy of HohaiGPDB could reach 59.38%,which proves the research value of Hohai-GPDB.The HohaiGPDB could be used as a general data set for protein-related studies.

Key words: Hohai graphic protein data bank, Protein spatial structure, Protein structure prediction, Transformer model