Recent technological advances have produced large data sets of protein-protein interactions (PPI). These data are often presented as graphs where nodes represent proteins and edges indicate the linkage between proteins with experimental support. This representation reveals that PPI networks are non-randomly organized, but unfortunately offers little insights on the dynamic nature of protein interactions inside the cell. The goal of this project is to develop generic methods for exploring the dynamics of PPI networks using human platelet cells as target and by integrating various information sources, with the eventual goal of being able to identify novel signaling pathways and networks in human platelets.
In this project a new graph model will be developed for capturing the dynamic nature of protein-protein interaction in the following manner. (a) Each node is not an individual protein but a protein-protein pair and each node is labeled with terms representing contexts in which the interaction should be observed. Each interaction node will be mapped to a real vector of fixed dimension using amino-acid sequence similarity and co-evolution as basis spectral clustering as a mapping tool. (b) Based on gene ontology, groups of interaction nodes will be presented as hyper-edges over the nodes. (c) The problem of inferring interaction will be translated as the problem of assigning labels to each node. (d) The label assignment problem will be solved as a label-ranking problem combined with a label number predicting problem. (e) The method will be validated using wet-lab experiments with human cells and human platelet data. Computational tools will be developed using the high-performance computing facility in the Center for Computational Science at the University of Miami. Undergraduate and high school students from the underrepresented groups will be actively recruited. Courses on computational biology and on chemical biology will be developed and taught at the University.