Selecting Near-native Protein Structures From Predicted Decoy Sets Using Ordered Graphlet Degree Similarity

With the completion of the genome project of humans and several other species,gene maps have been formally drawn,and life science has entered the post-genome era?Researchers want to understand the functions of hundreds of millions of gene sequences.It can be known from the genetics center rule that we can study the function of the corresponding proteins produced by gene expression.The function of a protein is known to be primarily determined by its tertiary structure.Therefore,the study of protein tertiary structure is crucial for understanding genomic data.However,only a small fraction of proteins currently have a known tertiary structure.Therefore,how to effectively predict the tertiary structure of proteins has become an important and challenging problem in computational structural biology.Using the ab initio method,a large number of candidate protein structures called prediction sets can be predicted,but choosing a good near-native protein structure from the predicted prediction set is a problem.In this work,we propose a new method for selecting near-native protein structures from the prediction set based on the similarity of contact map overlap(CMO)and graphlet nodal degrees.By extending the graphlet to an ordered graph and using dynamic programming to select the best alignment with the introduced dynamic gap penalty,GR_score is defined to calculate the similarity between the 3D protein structures.Finally,an integrated clustering approach was used to select near-native protein structures in the prediction set,and the centroid structure of the clusters was selected as a near-native protein structure.Experiments have shown that compared to the SPICKER method used in I-TASSER,the proposed method generally selects a better near-native structure based on the similarity between the selected structure and the native structure.