Scalable Graph - Mining Techniques with Applications to Systems Biology

SCHMIDT, MATTHEW C. Scalable Graph-Mining Techniques with Applications to Systems Biology. (Under the direction of Nagiza F. Samatova.) Genetic engineers often seek to modify the genome of prokaryotic organisms in order to improve their efficiency in certain industrial processes. This requires an understanding of the biological systems that are responsible for the expression of the organism’s physical traits, or phenotypes, that are required by the given industrial process. This thesis approaches the problem of predicting phenotype-related biological systems by searching for network motifs that model the biological systems, which are primarily present in the biological networks associated with phenotype-expressing organisms. This thesis consists of three components that enable the prediction of phenotype-related biological systems through the described approach. The first component is a framework that reduces the complexity of identifying phenotype-related metabolic systems through a simplified model of organism-specific metabolic networks and an efficient search method. The second component is a framework that identifies phenotype-related protein functional modules by efficiently searching for a specific subset of cliques in a organism-specific functional association networks. Empirical results show that these frameworks identify subgraphs that model systems that are highly biologically relevant. The third component is a scalable, parallel maximal clique enumeration algorithm that allows applications, such as that of the second component, to enumerate maximal cliques in large scale biological networks. Runs of this parallel enumeration algorithm on large-scale networks demonstrate that the algorithm’s runtime scales linearly to thousands of processors. c © Copyright 2011 by Matthew C. Schmidt