The central theme of this dissertation is predictive modeling of metabolism in

complex biological systems with genome-scale stoichiometric metabolic models to (i)

gain nontrivial insight on cellular metabolism and (ii) provide justifications for

hitherto unexplained metabolic phenomena. The crux of high-quality predictive

modeling with genome-scale stoichiometric metabolic models is appropriate selection

of (i) a biologically relevant objective function and (ii) a set of constraints based on

experimental data. However, in many complex systems, like a plant tissue with its

wide array of specialized cells, a biological objective is not always apparent.

Additionally, generation of experimental data to develop biochemically relevant

constraints can be nearly impossible in systems that cannot be cultured under a

controlled environment for the duration of an experiment. Such limitations necessitate

careful reformulation of the biological question, development of novel methods and

data analysis strategies.

Here, we push the boundaries of predictive modeling by demonstrating its first

application in deciphering hitherto unexplained metabolic phenomena and in

developing novel hypotheses on metabolism. Towards achieving this goal, we

developed several novel approaches and employed them in diverse biological

systems. Firstly, we investigated the selection of carbohydrate degrading pathway

employed by Saccharophagus degradans, an aerobic cellulosic marine bacterium.

Flux balance analyses of its growth in nutrient rich hypoxic marine environment

predicted that the selection of carbohydrate degrading pathway is possibly influenced

by inorganic nutrient availability. Secondly, multi-tissue genome-scale metabolic

modeling of Populus trichoparpa, a perennial woody tree, and analyses with a novel

strategy based on multiple biologically relevant metrics provided a metabolic

justification for the predominance of glutamine as the predominant nitrogen transport

amino acid for internal nitrogen recycling. Thirdly, predictive modeling of maize

grain filling predicted amino acid fermentation as a mechanism for expending excess

reductant cofactors for continual starch synthesis in the hypoxic environment of

endosperm. Finally, we developed bilevel optimization framework to integrate

publicly available transcriptome datasets with metabolic networks. This framework

predicted accumulation of specific classes of maize endosperm storage protein at

distinct stages of grain filling. We anticipate that the employment of these

aforementioned approaches in other biological systems will lead to the generation of a

wide array of nontrivial hypothesis on cellular metabolism and to develop targeted

experiments to validate them.