MAPK SIGNALING PATHWAY ANALYSIS ESD.342 FINAL PROJECT REPORT – MAY 16, 2006 Gergana Bounova, Michael Hanowsky, Nandan Sudarsanam I. INTRODUCTION Various biological systems have been represented as networks, and a significant subset of such representations is seen in cellular biology. Three distinct sets of network studies can be made in this topic, namely metabolic pathway networks, protein interaction networks, and genetic regulatory networks. Our study looks at a protein interaction networks and metabolic pathways. We have analyzed protein datasets for three different species from two different data sources. We have attempted to quantify regularities and trends, in the form of basic statistics and community structures. Comparisons are made across the pathways of the three species, as well as their equivalent random networks. A topic of core interest in this study was the application of a motifs analysis to the protein network. The study of structural and functional behavior of all proteins in the human body is termed as proteomics. It is a large scale ongoing project which is seen as a successor to the human genome project. Such an initiative is seen as being more complicated than the genome project since the estimated number of proteins is in the order of 400,000 as compared to 22,000 genes. Hence, it is clear that the various current constructs of the human protein network are at a much coarser level and possibly incomplete in many regions. The protein networks for simpler organisms have been studied to a greater detail, such as the Saccharomyces cervisae (baker’s yeast). Data sources for protein networks vary extensively despite bearing the same label. A model for the same network is constructed with as many as hundreds to thousands of nodes. This is based on the fact that the methods used to determine protein-protein interactions are based on different techniques and different motives for aggregating the data. A commonly used method known as two-hybrid screening is known to be able to successfully identify interactions within the same functional classes, while tending to ignore connections across functional classes. It is interesting to also note that the reliability that authors express about the data seems to decrease with networks that have higher numbers of nodes. The largest network that we came across was studied by S. Wuchty and E. Almaas, in their 2004 paper “Peeling the yeast protein network”, with 3677 nodes. In their paper they describe their data source as being “extensively flawed”. Another reason to further emphasize the inaptness of modeling protein networks is based on understanding the true nature of proteins and links. In contrast, genome regulatory links are structurally constant. Protein structure differs from cell to cell and is constantly changing based on interactions with the environment and the genome. Hence, even within an identifiable single cell, the protein structure might change through the life cycle of the organism. MAPK Signaling Pathways TeamThe data analyzed in this study is based on two sources. The KEGG (Kyoto Encyclopedia of Genes and Genomes) database is an initiative for constructing a complete computerized representation of the cell during what is termed as the post-genomic era. The other data source is the DIP ( Database of Interacting Proteins), which catalogs experimentally determined interactions between proteins. It combines information from a variety of different sources to provide a single consistent set of protein-protein interactions. Based on preliminary statistical analysis of the various networks, it was found that betweenness centrality was metric of key interest. It was found that when random networks (that preserved the same degree sequence) were created for each of the three protein pathways, this was the only metric that was significantly different in the original network. This regularity is seen in all three network pathways. Such a configuration is believed to be a sign of flexibility in the use of specific protein complexes and signal pathways for multiple different functions. A motifs analysis is carried out on the network. Typically, such an analysis is carried out by means of coarse graining the given network. Coarse graining is an important bottom-up method of understanding network structure, by uncovering global patterns (motifs). This helps us go beyond the global features and understand the relevance of certain structural elements. Motifs are statistically significant patterns of connections that recur throughout the network. These patterns serve as the building blocks for the network. Studies have shown that motifs identified in biological networks typically have certain key information processing function. II. LITERATURE REVIEW The use of network tools to represent biological systems is prevalent in literature, especially at the cellular level biology. A very broad paper summarizing the role and current application of network analysis to cellular level biological systems can be found in the work by Barabasi and Oltwar (2004). They provide discussions on the use of basic statistics, motifs, modularization and hierarchy, and their relevance to functional biology. In this paper, however, this study concerns only large statistics of the networks concerning protein pathways and interactions. A body of literature in this field focuses on constructing and establishing the network by means of various controlled experiments. Examples of such studies include Mansfield et. al. (2000) and Ito et. al.(2001). These papers do not perform any network analysis, but were used to identify data sources and understand the mechanism by which networks are created, predominantly, the two-hybrid screening approach. This technique is based on testing individual pairs of proteins for physical interactions (such as binding) by introducing genetically engineered strains of a certain protein construct. MAPK Signaling Pathways TeamThe second body of literature in this topic studies statistical properties and regularities in protein networks. Jeong et. al., (2001), look at the basic statistics of the protein interaction network found in yeast. Centrality is addressed in their study and a hypothesis that the most central proteins are the most important for the cells functioning is stated. They show that the network exhibits a scale free topology that is very similar to metabolic networks, and in general to that of robust