Abstract:Protein biological function is determined by its three-dimensional structure. Any residue substitution may lead to dramatic changes in structure and result in functional alternations. Thus it is of great importance to study the relationship between substitution and structure. Usually, proteins three-dimensional structure prediction is based on the similarity of sequence and structure. Whether this method is valid to the protein with residue substitution is still an open question. To answer this question, single residue substitution crystal structures were extracted from Protein Data Bank (PDB) to construct a non-redundant test set, which was used to assess two homology modeling structure prediction tools, SWISS-MODEL and MODELLER. The results indicate that the two methods perform well in overall structure prediction (RMSD<0.5Å), however, they fail to predict the mutants that own significant structural changes (A_RMSD>1.5Å) upon residue substitutions. The two methods perform better in mutations of embedded residues and mutations between polar residues. Although it is a small fraction of residue substitutions (<5%) that lead to significant conformation changes, they may play important roles in loss or gain protein functions. Therefore traditional homologous modeling methods are not appropriate for predicting such protein structures. There is urgent need to developing more powerful prediction methods.