Text

theses

Studying changes in the cellular metabolism is important to understand what a living cell does for survival in response to external or internal perturbations. Even though intracellular metabolic flux (i.e. reaction rate) distributions are desirable data to this end, it is challenging to directly quantify fluxes through methods such as metabolic flux analysis using stable isotope labeling. Several computational methods thus have been developed to infer system-level and condition-specific intracellular metabolic flux distributions, which are difficult to measure, from transcriptomic data, which are far easier to obtain. While powerful in many settings, existing methods have several practical shortcomings, and it is unclear which method has the best accuracy in general due to limited validation against experimentally measured fluxes. In this thesis, we describe two computational methods called E-Flux2 (E-Flux method combined with minimization of l2 norm) and SPOT (Simplified Pearson cOrrelation with Transcriptomic data), to be employed when a suitable biological objective is available and unavailable, respectively. Our method overcomes shortcomings of existing methods and combines desirable characteristics including applicability to a wide range of experimental conditions, production of a unique solution, fast running time, and the availability of a user-friendly implementation (at http://most.ccib.rutgers.edu/). Most importantly, the predictive accuracy of our method was validated using the largest experimental dataset compiled to date, consisting of 43 experimental conditions of transcriptome measurements coupled with corresponding central carbon metabolic intracellular flux measurements (19 in Escherichia coli, 9 in Saccharomyces cerevisiae, 8 in Bacillus subtilis, 3 in Synechocystis sp. PCC 6803, 2 in Synechococcus sp. PCC 7002, and 2 in H4IIE rat hepatoma cell line). Our method provided as good as or better predictions than a representative sample of competing methods including pFBA (parsimonious flux balance analysis), in terms of the average of correlation between predicted and measured fluxes and of overall stability in predictions, especially in unicellular heterotrophic microorganisms. This makes our methods useful even in the absence of measured flux rates that allow some existing methods such as pFBA to be employed. The goal of developing these computational tools is to better understand complex biological systems. Not only do the methods we developed contribute to advancing previous work, they have helped to answer biological research questions as well. In several collaborative research, our methods were used to understand the lipid accumulation mechanism of nitrogen-stressed Phaeodactylum tricornutum cells, verify the predictive power of a genome-scale metabolic model of the cyanobacterium Synechococcus sp. PCC 7002, and examine the metabolic impacts of RpiRc, a potent repressor of microbial toxins in Staphylococcus aureus.

ETD_8143

Ph.D.

Includes bibliographical references

by Min Kyung Kim

doi:10.7282/T39889SQ

ETD doctoral

The author owns the copyright to this work.