Skeleton-based Shape Similarity Analysis Of Three Dimensional Protein Models

In the post-genomic era, shape similarity comparison of three dimensional protein modelshas become an important research content in the protein shape analysis. The study on proteinshape comparison can reveal the structure and function of proteins. So it plays an important rolein structure-based drug discovery and protein structural retrieval and prediction etc. Because the3D protein model surface shape is complex and model deformation has an impact on thetopology change and similarity analysis. These problems lead to a great challenge on shapesimilarity analysis of proteins. This paper takes into full account the global and local features tocompare protein model similarity and does the following works.1. It analyzes the research status about the three-dimensional model skeleton extraction andexpounds the shape similarity analysis method of3D models. It also points out the importantresearch significance and wide application background.2. It proposes a shape similarity comparison of three dimension protein models (meshmodel and CPK model) based on their skeletons. And it uses the similarity function and greyrelation analysis to compare three dimensional structure of proteins.3. For a mesh model of proteins, it proposes an improved MRG skeleton exaction algorithmsolving the skeleton extraction problems based on the traditional MRG algorithm; For a CPKmodel, since there is no edge and facet relation, the MRG algorithm cannot be used to extract theskeleton of CPK models. So it proposes an improved L1-medial skeleton exaction process andgets a satisfactory skeleton result as the global feature.4. Then it uses the skeleton to construct local features of similarity analysis. It proposes aconstruction method of local features based on local radius or local diameter. Firstly, it evenlyextracts some points of a protein model as sample points. This can reduce the computationcomplexity which does not use all points of a protein model. And then it calculates the shortestdistance from the sample point to the skeleton of a protein model and forms one dimensionalvector as local radius feature.5. Finally it analyzes the similarity using the local diameter and local radius feature vectorsof protein models based on the distance measurement and grey relational analysis method. Itmainly uses the relative error distance and grey relational analysis method to analyze the similarity of protein models. And it compares the proposed method to other common methods,e.g. CE, Dali-Light, and FATCAT and so on. Experimental results show that the proposedmethod is more accurate than other popular methods.In the end of this thesis, it summarizes the main content of current research, and put forwardsome issues as the future work.