Three-dimensional protein shape similarity analysis based on hybrid features

Abstract

The analysis of protein similarity is a matter of concern in the bioinformatics field, since studying the protein similarity can help understand the protein structure-function relationship. To this aim, several methods have been proposed, but currently, protein similarity results are still not satisfactory. Here we presented a novel method for evaluating the similarity of 3D protein models based on hybrid features, including the local diameter (LD), the salient geometric feature (SGF) and the heat kernel signature (HKS). LD is suitable to the topological deformation of 3D models, SGF is an important local feature on the protein model surface, and HKS is invariant under isometric deformations. Our method provides the improved feature extraction procedure to calculate LD, SGF and HKS of a protein model, and then uses these features to construct a tensor based feature descriptor for 3D protein models. The method finally analyzes the similarity of 3D protein models by using this tensor descriptor and the extended grey relation analysis. Experimental data indicated that our method is effective and can outperform the existing similarity analysis results obtained by previously reported methods.