PhD Defense: Computational Methods for Pathway Synthesis and Strain Optimization
Engineering and optimization of biological cells have been central to modern biotechnology, with applications ranging from drug discovery and development to production of commercially significant chemicals. Purely empirical approaches can benefit greatly when paired with computer-aided design methods that allow for design space exploration and optimization. Such methods may significantly contribute to reducing experimental efforts in expediting discoveries.
This thesis addresses two problems in metabolic engineering and synthetic biology. The first problem concerns the construction of synthetic pathways to produce a desired metabolite within a microbial cell. We present two approaches for solving this problem. The first approach, ProbPath, identifies non-native de novo synthesis pathways from reactions within a database by probabilistically sampling available reactions. ProbPath is shown effective in identifying synthesis pathways when compared to exhaustive exploration of the design space with limited path length in terms of generating similar yield profiles. Additionally, we were able with ProbPath to reproduce routes that were experimentally obtained for the production of several molecules. The second approach addresses the issue when a desired target metabolite is not present in known databases. To produce a synthesis pathway or such a metabolite, we develop a novel methodology based on identifying structural similarities between the target metabolite and existing metabolites within the database and developing transformation operators that predict the transformation outcome when applied to the target metabolite. To study this approach, we developed an algorithm, PROXIMAL, to construct transformation operators based on the set of xenobiotic transformations associated with human liver enzymes. We evaluated the prediction accuracy of PROXIMAL through case studies on two environmental chemicals. Comparisons with published reports confirm that our predictions have been experimentally validated in the literature.
The second problem addressed in this thesis concerns identifying optimal gene modifications when tuning a microbial cell to maximize the production of a desired compound. The novelty in our problem formulation lies in explicitly accounting for likely variations in flux capacities due to engineering modifications. The thesis presents a computational framework, CCOpt, which identifies an optimal set of gene modifications. CCOpt is based on chance-constrained programming, where constraints are probabilistically met at a user-specified confidence level. Evaluation of the approach demonstrates that CCOpt consistently finds a solution most-frequently found when using Monte Carlo sampling, but at a fraction of a computational cost. The CCOpt formulation is the first work to incorporate uncertainties when computing gene modifications.
Overall, the thesis contributes and advances the state-of-the-art in design automation tools for metabolic engineering and synthetic biology.