| Single point mutations, leading to missense mutants, are analyzed using a computational geometry approach, based on the Delaunay tessellation of proteins, and evolutionary-based methods. Missense mutants are of particular interest because they may cause structural alterations of the protein and lead to a loss of function. To test our computational approaches, we used a large dataset of 210 protein structure pairs and then narrowed our analysis to two DNA binding proteins (the bacteriophage Gene-V-protein and the human tumor suppressor protein p53), for which ample structural and experimental functional data are available. First, we show that statistical scores calculated for missense mutants and derived from the Delaunay tessellation of proteins (which gives a "map" of all four nearest-neighbor residues or quadruplets), correlate well with prominent structural features, as well as with stability and function. Second, a method using potential score differences, which measure changes in amino acid composition of quadruplets between wild-type and mutant proteins, was developed to predict the functional impact of missense mutants in the DNA binding domain of p53. This novel method yielded a prediction accuracy as high as 78.5%. Third, an evolutionary-based method, A-GVGD, based on multiple sequence alignments was tested for its ability to predict the function of p53 missense mutants. Compared with two other sequence-based methods, SIFT and Dayhoff's classifications, A-GVGD gave the best results with prediction accuracies as high as 59.3% for functional mutants and 88.3% for non-functional mutants. Finally, the occurrence of mutations in human cancers, estimated from their frequencies in the IARC TP53 database, was analyzed in light of dinucleotide mutation rates and functional activity of mutants, to evaluate the impact of mutagenesis and biological selection on cancer specific mutation patterns. Overall, our results provide conclusive evidence that the structural and functional impact of single point mutations may be predicted using both the computational geometry and A-GVGD methods. |
