Research On Prediction Algorithm Of Protein Discontinuous Domain Based On Graph

As an important part of protein three-dimensional structure,domain has a direct impact on the function of protein.Accurate recognition of protein domain is very important for protein structure analysis.Although the traditional method is accurate and reliable,it has high cost and low efficiency.The prediction of protein discontinuity domain is helpful to the determination of protein tertiary structure and the study of function,and is of great significance to the understanding of disease mechanism and the development of new drugs.The research methods based on sequence seldom consider the detection of discontinuous domains.Although most of the methods based on structure can consider the boundary division of discontinuous domains,they are relatively dependent on the three-dimensional structure of proteins.Different from previous research ideas,the method of protein discontinuity domain prediction based on graph starts from sequence,combines graph correlation method to predict the boundary of structural domain,and finally detects discontinuous structural domain.Research ideas: firstly,starting from the sequence,combining with deep learning algorithm to predict the residue contact map,then determining the protein contact graph based on the residue contact map.Two algorithms are proposed for recursive segmentation of protein contact graphs to obtain domain boundary information.Finally,discontinuous structural domains are detected by structural domain detection algorithm.The first segmentation algorithm is based on the maximum flow method,which decomposes the graph recursively until the subgraph after decomposes meets the termination condition and the algorithm ends.The second algorithm is based on fiedler vector method,this method with the first method process is roughly same,the difference is that the method is calculated Laplace time small Laplace eigenvalue and fiedler vector,in every division,need according to the result of fiedler vector-valued traversal segmentation,select the optimal segmentation result for the current decomposition results.The results show that the fiedler method is better than the max-flow method in accuracy,recall rate,NDO-score and DBD-score,and better than the ThreaDom method in accuracy,recall rate and NDO-score.Therefore,it can be proved that the protein discontinuous domain prediction method based on graph is reasonable and feasible.