Open Access
Kumar, Vinay
Graduate Program:
Industrial Engineering
Doctor of Philosophy
Document Type:
Date of Defense:
March 04, 2010
Committee Members:
  • Dr Costas D Maranas/ Dr Soundar T Kumara, Dissertation Advisor
  • Costas D Maranas, Committee Chair
  • Soundar T Kumara, Committee Chair
  • James Gregory Ferry, Committee Member
  • David Arthur Nembhard, Committee Member
  • Arunachalam Ravindran, Committee Member
  • Optimization
  • systems biology
  • metabolism
  • gaps
The central theme of this dissertation is the development of computational frameworks to analyze and improve the content of metabolic networks. The key questions that are addressed in this work are how does one a) characterize and automatically identify topological inconsistencies in a given metabolic network, b) rectify these inconsistencies by modifying these networks both internally and externally, c) develop a systematic framework to compare predictions of a given metabolic network with observed experimental data, d) develop optimization tools to reconcile any inconsistencies between model predictions and observed data and e) apply these tools during the elucidation of metabolic networks for novel organisms instead of applying them to existing metabolic networks. In the first part of the dissertation, we will provide the background in biology needed to understand the role of metabolism in the cell. We will then introduce the formalisms established to model this cellular process as a large-scale network. Equipped with these concepts, in the second part of the dissertation, we will characterize notions of topological inconsistencies (“gaps”) in these networks and design a Mixed Integer Linear Program (MILP) (GapFind) developed to automatically pinpoint them. Subsequently, we will introduce a MILP (GapFill) developed to automatically correct these inconsistencies by minimally adding foreign reactions from external databases and alternatively, by suggesting modifications of thermodynamic assumptions in the existing network. We demonstrate these procedures by analyzing the metabolic networks of Escherichia coli and Saccharomyces cervisiae (commonly known as yeast), which are model organisms used extensively in the biological community. In the next part of the dissertation, we utilize the gold standard for testing the accuracy and completeness of these models of metabolism, which is to compare their cellular growth predictions (i.e., cell life/death) across different scenarios with available experimental data. Although these comparisons have been used to suggest model modifications in prior efforts, the key step of identifying these modifications has often been performed manually. In this part of the thesis, we describe an automated procedure GrowMatch that addresses this challenge. When the model over-predicts the metabolic capabilities of the organism by predicting growth in contrast with experimental data, we use a minimax inspired bilevel optimization model to restore consistency by suppressing growth enabling biotransformations in the model. Alternatively, when the model under-predicts the metabolic capabilities of the organism by predicting no growth (i.e., cell death) in contrast with available data, we use an MILP to restore consistency by adding growth-enabling biotransformations to the model. We demonstrate the use of GrowMatch by reconciling growth prediction inconsistencies of the latest Escherichia coli model with data available at the Keio database. In the final part of the dissertation, we will demonstrate the use of these procedures during the reconstruction of metabolic models of the organisms, Methanosarcina acetivorans and Mycoplasma genitalium.