Network refinement (Fig. 10.3c) can be viewed as reconciliation between the content of the model and the available biochemical and genomic data, with the end result of enhancing the reconstructed network. This reconciliation is done based on agreements of model simulations and updated genomics, physiological, and bio­chemical knowledge. A crucial step in the reconstruction of genome-scale meta­bolic models is filling the gaps to decrease the number of dead-end metabolites and improve network connectivity. Metabolic network gaps are filled by the addition of reactions that are missing in the network yet have corroborating evidence for their existence in the system. These may include spontaneous reactions that are not associated with gene products as well as extracellular and intracellular transport reactions and exchange reactions.

Models may not predict the production of biological compounds with existing biochemical evidence if the prerequisite genes have not been added to the model. Manichaikul et al. (2009), using Chlamydomonas as a model, described how genomic data can be used to fill gaps in metabolic models. In their approach, not only genomics and other experimental evidence contributed to the refinement of the network, but also the model itself informed “genomics,” of the presence of missing annotations, justifying the use of more sensitive sequence search and annotation tools to recover the missing genes. One example that can illustrate this is lactate dehydrogenase (LDH), which initially was absent from the Chlamydomonas gene annotation, yet the model reconstruction showed the need for the LDH enzyme in the Chlamydomonas pyruvate metabolism pathway. A PSI-BLAST analysis was carried out to identify the gene encoding LDH; the gene was subsequently added to the model. Additionally, orphan genes, or those biochemically characterized metabolic enzymes lacking sequence data, can be assigned GPRs by reviewing metagenomic sequence data to provide sequences for the missing enzymes. This approach has been experimentally validated (Yamada et al. 2012).