Protein Structure Classification Based on Chaos Game Representation and Multifractal Analysis

Classification of protein structures is important in the prediction of the tertiary structures of proteins. In this paper, we propose to decompose the chaos game representation of proteins in to two time series, from which the protein sequences can be uniquely reconstructed. Multifractal analysis is applied to measures constructed from these two time series. A total of 26 characteristic parameters are calculated for each protein, which are used to construct a 26-dimensional space. Each protein is represented by one point in this space. A procedure is proposed to classify the structures of 100 large proteins consisting of four structural classes. Fisher's linear discriminant algorithmdemonstrates that the average accuracy for our classification can reach 84.67%. Compared with the results for the 46 large proteins reported before, the method proposed here has much better performance.