Increasing consensus of context-specific metabolic models by integrating data-inferred cell functions

Genome-scale metabolic models provide a valuable context for analyzing data from diverse high-throughput experimental techniques. Models can quantify the activities of diverse pathways and cellular functions. Since some metabolic reactions are only catalyzed in specific environments, several algorithms exist that build context-specific models. However, these methods make differing assumptions that influence the content and associated predictive capacity of resulting models, such that model content varies more due to methods used than cell types. Here we overcome this problem with a novel framework for inferring the metabolic functions of a cell before model construction. For this, we curated a list of metabolic tasks and developed a framework to infer the activity of these functionalities from transcriptomic data. We protected the data-inferred tasks during the implementation of diverse context-specific model extraction algorithms for 44 cancer cell lines. We show that the protection of data-inferred metabolic tasks decreases the variability of models across extraction methods. Furthermore, resulting models better capture the actual biological variability across cell lines. This study highlights the potential of using biological knowledge, inferred from omics data, to obtain a better consensus between existing extraction algorithms. It further provides guidelines for the development of the next-generation of data contextualization methods.Author summary: Genome-scale models of human metabolism have facilitated numerous exciting discoveries regarding human physiology and therapeutics. The accuracy of results from such studies requires that models capture the tissue or cell-type specific metabolism. In hopes to obtain accurate models, several algorithms have been developed to extract cell- or tissue-specific metabolic models. Each algorithm has provided useful insights into the metabolism of specific cell and tissue types. However, since each of these methods use different assumptions to guide reaction inclusion and removal, they result in considerable differences in size, functionality, accuracy, and ultimate biological interpretation, even when using the same data set. To overcome this, the enclosed research proposes an approach to infer the functionalities of a cell or tissue from omics data, and then protect these functions to guide the construction of a context-specific model. Through this study, we highlight the value of using experimental data to help infer the set of metabolic functions that should be included in a model, in an effort to obtain greater consensus across existing extraction algorithms. This study further provides guidelines for the development of the next-generation of data contextualization methods.