Some Applications Of Fractal And Network Methods To Protein Data Analysis

Gene is the carrier of genetic information, the proteins encoded by genes are the direct manifestation of the complexity and variability of life phenomenon, so the study of protein is signi?cant. This thesis mainly studies the related data of proteins. The arrangement of this thesis is as follows:In Chapter 2, we study the prediction problem of disease genes. In recent years, the researchers presented many kinds of algorithms to rank candidate genes from a single data source to integrate multiple data sources. We also use integrating multiple data sources. We propose the Laplacian normalization and random walk on heterogeneous networks to predict disease genes. The results show that our algorithms have superior performance.In Chapter 3, we study the prediction problem of protein folding type. Protein folding process focuses on two aspects: protein folding rate and folding type, here we study protein folding type. Previous methods for predicting protein folding type require to use the information on tertiary or predicted secondary structure of a protein. Here, based on physicochemical properties of amino acids and sequence itself, we extract various characteristics to predict the protein folding type from the primary structure of a protein using support vector machine combined with principal component analysis. The results indicate that the present approach is effective and valuable.In Chapter 4, we study the prediction problem of low-homology protein structural classes. In recent years, research shows that one can effectively predict protein structural classes based on the predicted secondary structures of proteins. Here,the chaos game representation is employed to represent the predicted secondary structure,from which we construct visibility network, horizontal visibility network and extract network features. We use support vector machine to predict the structural class of each protein. The results show that network features also have the obvious classi?cation effect.In Chapter 5, we study the fractal and network analysis of two-state and multi-state proteins based on the molecular dynamics simulation. Here, we perform molecular dynamics simulations for two-state and multi-state proteins, and extract various energy time series to construct horizontal visibility network and signal network. We calculated the fractal dimension of horizontal visibility network, ?tted the distributions of out-degree and in-degree of symbolic network respectively, and the results show that those distributions follow the inverse gaussian distribution.