Abstract
Genome-Scale Metabolic Models have shown promising results in biomedical applications, such as understanding cancer metabolism and drug discovery. However, to take full advantage of these models there is the need to address the representation and simulation of the metabolic phenotypes of distinct cell types. With this aim, several algorithms have been recently proposed to reconstruct tissue-specific metabolic models based on available data. Here, the most promising were implemented and used to reconstruct models for two case studies, using omics data from distinct sources. The set of obtained models were compared and analyzed, being shown they are highly variable and that no combination of algorithm and data source can achieve models with acceptable phenotype predictions. We propose an algorithm to achieve a consensus model from the set of models available for a given tissue/cell line, and to improve it given functional data (e.g. known metabolic tasks). The results show that the resulting models are more accurate, both considering the prediction of known metabolic phenotypes and of experimental data not used in the model construction. Two case studies used for model validation consider healthy hepatocytes and a glioblastoma cell line. The open-source implementation of the algorithms is provided, together with the models built, in a software container, allowing full reproducibility, and representing by itself a contribution for the community.