The molecular basis of cell phenotypes can be elucidated through the use of genome-scale models of metabolism. However, especially for higher organisms, many pathways and enzymes are used only under specific conditions. To enhance the predictability of metabolic network models for cell lines and tissues, several algorithms have been developed that allow one to tailor a genome-scale model to a specific context by incorporating omics data. However, these methods have not been benchmarked, and it is unclear how choices of algorithm or parameter selection ultimately impact the resulting model content or predictive power of these models. To investigate this problem, we implemented several leading algorithms with a variety of parameters to develop hundreds of models of the A375 melanoma and HL60 leukemia cell lines. In a thorough analysis of these models, we demonstrate that pathways included in a model vary substantially based on a few key decisions in model construction. Despite the variability in model content, the extracted models successfully increase the accuracy of model-predicted gene essentiality, and the models recapitulate many known metabolic functions of the cells. We further enumerate the key decisions that most significantly impact model content, and gain insights algorithm assumptions that ultimately influence the ability for the resulting models to recapitulate actual cell physiology. Thus, as these insights are used to guide the development of cell line and tissue-specific models, predictive accuracy of metabolic network models will continue to increase.

Hierarchy of decisions that impact the accuracy of tissue-specific metabolic models

251st National Meeting (2016)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.