A number of genome-scale networks have been reconstructed based on available data for the target organism. At present, however, there is no organism for which a complete data set exists. GEM-derived predictions based on incomplete reconstructions will fail when the expressed phenotype relies on a missing component or interaction. The mismatch between prediction and observation can be used to build hypotheses about the missing components. Given the high-dimensional models and data sets in genome-scale science, computer algorithms facilitate hypothesis generation greatly [297, 478].

Models Can Drive Discovery

Missing information As detailed in Chapter 3, metabolic models are constructed from a biochemically, genetically, and genomically structured knowledge base. This knowledge base couples the components of a metabolic reaction with its catalytic enzyme and the gene(s) that encodes it. The information used to create the knowledge base is incomplete, and Figure 25.1 summarizes how we might be missing information. A reconstruction can have missing reactions, protein, or genes. This chapter discusses how genome-scale models can be used to discover systematically some of the missing information using a combination of computation and experimentation.

The iterative nature of high-dimensional model-building Network reconstructions represent a knowledge base about the target organism that leads to the construction of predictive models of organism function. Because, at any given point in time, the information available for the target organism is incomplete, the genome-scale model cannot compute all organism functions correctly. Some of the model-derived predictions will be inconsistent with the data obtained and some will be consistent (see step two in Figure 25.1G). The consistent computational outcomes can be considered validations of the GEM (Chapter 4), while the inconsistent ones represent failure of prediction. The analysis of these failures leads to model updates and a re-versioning of the model. The model-building process is therefore iterative and constantly incorporates and reconciles more information.