Approach To Prediction Of Protein-Protein Interaction Interface And Hot Spot

Protein-protein interaction is the hot and difficult spots in the molecular biology investigation. Activities of the protein in cellular levels are executed by protein-protein interactions, such as cellular metabolism, signal transduction, immune recognition, DNA replication, gene translation and protein synthesis. The study of protein-protein interaction interface and its diverse structure and special structure recognition which can coordinate the function in complex is an important way to reveal the nature of life phenomenon in the molecular, cellular and organism levels.This thesis presents research of the interface characteristics in protein-protein interaction based on machine learning methods to study the interface domain in protein-protein interaction, and the key points in interface, and identify the structure module with biological function. It provides a new approach to explain protein function. The thesis found that interface's research potential in computer science and mathematics through analysis on the interface's research. This thesis is based on the above analysis to carry out research work. First, illustrating a new protein's correlation coefficient encoding based on the amino acid sequence to predict the interface. Second, proposing network topology construction method based on graph theory to calculate the complete interface domain. Thirdly, random integration of multiple features is proposed to predict the hot spot residues. Finally, preliminary research work about the identification of the hot region is a way to specify further research.The results about interface in this work performance are more stable, because a new protein's correlation coefficient based on the amino acid sequence is illustrated. The encoding sequence schemes consider the internal long-range interactions which can improve the protein encoding by maths model. The results of predicting hot spots show more desirable, which reflects the multi-feature random amalgamation (MFRA) that can break through the limitation of property strategy relying on parameter index. Moreover, MFRA can maximum probability of finding the best combination of features in limited possibilities. MFRA is not only quantitative analysis of feature contribution but also feedback learning. Thereby, this method can strength the prediction of hot spots and enhance the cognitive of hot spot on the protein's structure and function.