The Research Of Protein-protein Interaction Networks Based On The Wavelet

The rapid development of biological experimental techniques leads to vast amounts of experimental data. As the computing power of silico continues to improve, a new research platform is provided for high-speed processing of biological data. All of those factors have effectively promoted the study of bioinformatics in-depth development. How to mine useful information and summarize practical and significant laws from those vast amounts of biological data is one of goals of bioinformatics.Protein is an important sort of molecules in living being. The interactions between them support the activities in vivo, such as the structure of cell, the differentiation and growth of intracellular important organelles, the regulation of genetic information, the enzymes catalytic effort, the immune protection and so on. Therefore, knowing interactions in vivo well could contribute to a better understanding of the inner mechanism of life. For some practical problems, such as the cure for disease, it can provide some better alternatives. Though a flood of protein interaction data has been produced with progresses of experimental techniques, the existing data are still unable to explore the inner physiology of the life to provide strong support for such an ambitious project; because the inherent complexity of living being. It seems that this project can not be achieved by the continuous improvement of experimental techniques alone. In the other side, the vast amounts of data brought about by advances in developing technology methods based on data analysis (e.g., computing method). As computing power continues to improve, methods based on protein sequence information to analyze and predict protein-protein interaction have gradually become more and more important. Currently, the common methods is using background information and tools to extract the features of interacting proteins, and then use machine learning methods to learn those features for predicting protein-protein interactions and finanlly construct important interaction networks.The main content of this paper is to extract protein-protein interaction features in the term of spectral information. It includes the detailed introduction of protein spectral information method and ideas, the use of it combined with lifting wavelets to predict protein-protein interactions and finally construct interacioning networks.This article is divided into five parts: 1) From the original of the Bioinformatics, to the Genomics, then to the proteomics, a general description of the importance of the protein-protein interaction is given. On the base of it, most of past and current methods to probe and predict protein-protein interactions, besides obtaining results and their advantages and disadvantages are elaborated in detailed. Then we introduce 3 kinds of protein structure, how they have much effort on the proteins'function, where they are stored and how to construct the local database of those kinds of data to design and improve practical algorithm.2) For different structures of datasets or networks have great impact on performance measurement of methods, especially those based on sequence information. Based on Previous work, methods based on sequence information cannot accurately learn and then predict some protein-protein interaction in those special networks which contain some proteins with high degree (hub protein). In this chapter, we discuss different local interacting network has it own features. Based on calculating results, features of Protein-protein interactions belong to the same local interacting network can be caught in the use of the machine-learning methods and balance datasets. But features will be covered and cannot be caught, if there are some proteins with high degree. Besides, we also use the associate rules to find out what group of 3-amino acid make more contribution in thoselocal networks whose features can be caught.3) Resonant recognition model (RRM) is an important tool to analyze the nature of protein. It mainly gets the information of protein from the information spectrum. The application based on information spectrum is Extensive, such as the prediction of active sites of pretein, the prediction of protein second structure, the prediction of protein-protein interactions and protein-DNA interactions. In the chapter four, we firstly introduce the basic principle of RRM and its application. For the RRM cannot show the local information, the wavelet is imported. After describe the theory of wavelet, some meaningful application based on RRM and wavelet are also introduced, including those work of us, predicting the interaction between proteins and the interaction between protein and DNA.4) In the chapter five, we use the lifting wavelet to calculate the protein-protein interacting features from the protein's sequence. These features are learned by Support vector machine and then a model is trained by which we predict protein-protein interactions. Numerical experimental results report that, on the principle of balance between positive dataset and negative dataset, the low-dimensional vector of features has gained a better performance. Calculation results also report features of different functional interacting network are different, which cannot be detected only by one method. To make a more accuracy prediction, several different kinds of features are essential.5) The purpose of protein-protein prediction is to construct the interacting network. From the interacting network, one can have a deeply insight of living being, the relationship between interacting network and biological function in cell. Therefore, in the chapter six we introduced general methods to construct the interaction network of proteins. Then we choose several local interacting networks. We use the 3 amino-acid-frequency or ones mined by lifting wavelet of the protein pairs in one local interaction network, and then combined them with machine-learning method to build the predicting model. After the model is built, we choose some important protein pairs from database as ones waiting to be predicted. From predicting results, we construct the interacting network and plot the figure of interacting networks.