LAUSR

Metabolism is crucial for all living cells since it provides energy and molecules for all biological functions. Systematically understanding metabolism is therefore important both in medical research and in synthetic biology where it can help to better engineer cells. Over the last decade, researchers have built genome-scale metabolic models (GEMs) to systematically simulate the complete known metabolism of the organism of interest. However, there remain many knowledge gaps in these models such as unannotated and misannotated genes, promiscuous enzymes, unknown reactions and pathways, and underground metabolism. Gaining insight into the unexplored metabolism is essential because a detailed understanding of cellular functions drives biomedical applications such as drug-targeting strategies. On the other hand, exploring the knowledge gaps in metabolism is also necessary for synthetic biology. For example, knowing the complete synthetic pathways enables efficient design of cell factories to produce chemicals and pharmaceuticals, especially valuable secondary metabolites (1). Thus, in order to systematically identify and reconcile the metabolic gaps at a genome scale, Vayena et al. (2) propose a computational gap-filling workflow, Network Integrated Computational Explorer for Gap Annotation of Metabolism (NICEgame). Vayena et al. (2) applied this gap-filling workflow NICEgame to identify and reconcile the knowledge gaps in the latest Escherichia coli GEM iML1515 (3). They compared the model prediction and experimental phenotype of E. coli single-gene knockouts in glucose minimal media and identified metabolic gaps for 148 false gene essentiality predictions linked to 152 reactions. The workflow NICEgame can propose alternative reaction sets as gap-filling solutions to reconcile the false essential gene predictions. Earlier gap-filling methods rely on biochemical reaction databases or published GEMs as reaction pools for gap-filling (4). Solutions suggested by those algorithms are limited within the scope of known biochemical reactions. There may exist only a unique solution for filling the same gap, leading to much more identical biased metabolic networks among diverse organisms, especially for those organisms which are poorly annotated and contain a large part of knowledge gaps. The NICEgame workflow relies on a much more extensive reaction database, ATLAS of Biochemistry (5), consisting of known and broader hypothetical reactions built from mechanistic understandings of enzyme function mechanisms, which provides more possibilities for the knowledge gaps and enables the identification of new biochemical capabilities and enzyme functions, ensuring more knowledge gaps to be filled and more gap-filling solutions to be found. In the case study of E. coli, Vayena et al. (2) identified that the average number of solutions per rescued reaction is 252.5 when using ATLAS as the reaction pool versus 2.3 when using the KEGG reaction database, a resource covering known biochemical reactions. The comparison was performed by constraining both reaction pools within the scope of E. coli and yeast metabolites. Moreover, 53 of the total identified 152 false essential reactions were reconciled with thermodynamically feasible gap-filling solutions when using the KEGG reaction database as the gap-filling reaction pool. In comparison, 93 of 152 false essential reaction gaps can be rescued using the subset of ATLAS. Besides that, compared with the earlier gap-filling methods, NICEgame outputs alternative subsets, allowing users to evaluate those subsets based on biological domain knowledge (Fig. 1). With more subsets being proposed for each targeted gap, the question remains of which one should be added to the model. Vayena et al. (2) adopted a scoring system to rank the reaction subsets by considering the thermodynamic feasibility and the minimum impact on the model. The introduction of longer paths, new metabolites, and novel enzyme functions (when the third level EC number does not exist in the original GEM) was panelized. Annotating genes for the proposed reactions for gap-filling is extremely useful, which drives further experiments to identify novel metabolic discoveries. In the NICEgame workflow, Vayena et al. (2) adopted BridgIT (6), a previous tool developed in the Hatzimanikatis group, to identify the enzymes associated with reactions for gap-filling. Reactions annotated with enzymes of higher BridgIT confidence scores were favorable. In their paper published in this issue of PNAS, they proposed 77 new reactions associated with 35 E. coli genes to extend the latest E. coli GEM iML1515 to reconcile 47% of the 148 identified false essential gene predictions. Among these 35 genes, 33 were present in the original GEM iML1515. The added new reactions show the substrate or mechanism promiscuity of these 33 genes. Two new genes, ArcA and LacA, which were not part of the original reconstruction, have been assigned reactions and added to the model. In total, the added biochemistry reconciles metabolic gaps linked to the amino acid metabolism, cofactor metabolism, and biosynthesis of cell membrane peptidoglycans. The performance of the extended GEM of E. coli, iEcoMG1655, was validated on the gene essentiality experimental data on 15 carbon sources (3). iEcoMG1655 showed a 23.6% accuracy increase in gene essentiality predictions compared with the original GEM iML1515.