Grant: Discovery Grants Program - Individual

If Shakespeare were a Computer Scientist, he might say "all the world's a graph". We use graphs to represent information and relationships at many different levels: towards the upper end, interactions between people can be represented in a contact network, while towards the lower end there are protein-protein interaction networks. Regardless of what information is being represented, there are two main issues: how to obtain accurate graphs, and how to extract and infer meaningful information from them. My research program uses computational intelligence and combinatorial analysis to address these issues. Evolutionary algorithms are developed and analyzed for their success at generating graphs that meet specific criteria. These algorithms are then applied to real-world problems. For example, these algorithms are used to generate contact networks for which an epidemic unleashed upon the associated population would likely have certain properties, such as infecting many people. Combining this with a model to represent the creation of epidemic variants, we can extract useful information about the conditions that make variants successful, so as to help inform mitigation or vaccination efforts. Algorithms are also developed to analyze biological data stored in the form of graphs. Storage of such data is greatly complicated by both size and noise; graph compression reduces the size of the graph and simplifies its structure, making analysis more manageable. We develop and analyze a graph compression methodology that evolves the compression so as to minimize the distortion created by the compression. As this methodology is lossy, it can help to smooth out noise. Developing algorithms to analyze the information stored in complex networks that combine information from many types of interactions (between proteins, genes, etc.), we can predict genes that are associated with a given disease. This is incredibly important in an era in which there is a rapid increase in personal DNA testing and analysis, and it allows for the possibility of faster diagnosis and targeted therapies. Both genetic factors and observed symptoms are components in diagnosing disease. In analyzing data from Parkinson's disease patients, we can model aspects of the disease. There are several ways in which this can improve quality of life: it can be used to identify changes that signal the need for changes in treatment, and to predict and prevent freezing of gait. In all the proposed work there are several common themes: the use of computational intelligence, and the analysis of biological and biomedical data in the form of graphs. The techniques, their analysis, and their application, are of interest to computer scientists, while the results obtained from analyzing the biological and biomedical data are of interest to those studying the diseases, and to public health officials.