Prediction Research Of Protein-Protein Interaction Sites Based On Covering Algorithm

Protein is the primary Performer of biological function, It is the interaction between them that achieves the Organism's function. Researches on interaction between proteins and proteins are not only good for analyzing protien functions and understanding mechanism of life activities but also have important practical significance of diagnosing disease and pharmaceutical design. Traditional experimental methods of identifying protien-protien interaction are time-comsuming and labor-intensive and have constraints of science and technology, time and space which make the results contain high rates of both false positive and false negative predictions. In addition, the accuracy of prediction is also lower. So it is essential to seek efficient computional methods to identify protien-protien interaction sites. Between those computional methods, the methods of mechine learning have been widely used in predicting protien-protien interaction sites with good results. This paper adopts constructive algorithms for neural networks-covering algorithm, the kenel covering algorithm and Multi-side Increase by Degrees Algorithm to achieve the task of identifying those sites. It is achieved very good results. The main work of this thesis are summarized as follows:(1) Introducting the background, significance and status of the research. Summarizing the basic concepts of protien and experimental and computional methods predicting protein-protein interaction.(2) Describing covering algorithm and its improved algorithms and constructing predicting model to identify protein-protein interaction sites. Firstly, Summarizing geometric meaning of covering algorithm, highlighting the improved covering algorithm-kernel covering algorithm combining the advantages of covering algorithm and support vector machine which has such advantages as strong robustness and high precision. After that, the result was abtained by constructing model of prediction using the kenel covering algorithm. Discussing the result of the kenel covering algorithm and its superiority is presented by compraring its result with the results of BP algorithm and support vector machine. In addition, the contribution of predicion performance of residue accessible area was tested by introducing it based on two protien features of sequence profile and entropy.(3) The Multi-side Increase by Degrees Algorithm is a method which devides the problem into several parts and deals with those parts. Firstly, analyzing the problem from primary to secondary to get its basic features. Then, comprehensively analyzing it using those features. Thus, it can not only chosse effective dimensions and reduce dimensions to reduce computational complexity but also match complex problems from different aspects to improve its generalization ability. However, the task of predicting protein-protein interaction sites is just a high-dimensional and complexly computional problem. In this paper, the Multi-side Increase by Degrees Algorithm was firstly used as an attempt to identify protein-protein interaction sites. The result of the Multi-side Increase by Degrees Algorithm demonstrates the feasibility and effectiveness of prediction after comparing it with that of BP algorithm and SVM.