Figure 1.
A flow diagram summarizing the BioMog framework.
Initially, mutants are evaluated computationally to determine metabolites that are blocked in mutant strains but not in the wild type strain. While not strictly necessary, mutants that have substantially deleterious impacts on the network (>100 blocked metabolites) are considered uninformative and filtered out to improve the quality of proposed biomass components. Blocked metabolites are then assigned to an include/exclude metabolite list based on the experimentally observed growth phenotype. Once these lists have been created, BioMog can use this information to propose de novo biomass components or to modify the existing biomass equation. Additional experiments can be designed and run by BioMog to fill in any informational gaps that may exist in the current dataset after which the cycle can be repeated.
Figure 2.
Application of BioMog to an illustrative example.
(A) For an existing model and set of biomass requirements (metabolites F and H), BioMog is capable (depending on the quality and quantity of data) of generating, de novo, an organism's biomass requirements or of modifying a predefined biomass equation. This is accomplished by removing the initial biomass equation from the network and adding sinks for every metabolite (not shown). Blocked metabolites are determined for the wild type and mutant strains under a particular experimental condition (B). This process is repeated for all growth phenotype experiments for which data exist. The set of blocked metabolites can then be used to propose a new biomass equation or modify an existing one (C). Based on the include/exclude metabolite lists generated in this example, the original biomass equation composed of substrates F and H is modified by adding metabolite C while removing F. Since the de novo biomass relies solely on experimental evidence, it is important that enough data exist that test the essentiality of different metabolites if one desires an accurate and complete understanding of the biomass requirements. Here, metabolite H was absent from the proposed de novo biomass because no supporting or refuting evidence existed in the experimental dataset to justify its inclusion.
Table 1.
BioMog proposed de novo biomass equations for E. coli and S. oneidensis.
Figure 3.
Percent of growth phenotype agreements for the BioMog De Novo biomass and the predefined biomass.
Unsurprisingly, the BioMog biomass equation consistently outperforms the curated biomass over the entire dataset. Numbers reported above a given bar indicate the total number of experiments for that category.
Figure 4.
Frequency of appearance in the include/exclude metabolite lists.
Frequency of select predefined biomass metabolites in include (red bars, corresponding to experiments where mutants do not grow) and exclude (blue bars, corresponding to experiments where mutants grow) metabolite lists, as well as the difference between the two (green bars) for both E. coli (A) and S. oneidensis (B). For E. coli, 29 of the 72 predefined components (40%) had no direct supporting or refuting experimental evidence (i.e., the red and blue bars were zero). A positive difference between the include and exclude frequency indicates a potential improvement in agreement by including the metabolite as a biomass component. Note that a positive difference does not ensure that a metabolite's inclusion will lead to a maximal agreement score. Selection of biomass components from multiple blocked metabolites associated with a given experiment can potentially replicate the same experimental phenotypes while minimizing overall disagreements (see text for details). Red labeled metabolites indicate those that were recommended for removal by the biomass modification algorithm. Metabolite abbreviations match those used in the two metabolic models.
Figure 5.
Comparison of predefined biomass to BioMog's de novo (top) and modified biomass (bottom).
Numbers above bars indicate the number of metabolites in each category. iJO1366 and iSO783 are the genome-scale models used for E. coli and S. oneidensis, respectively.
Figure 6.
Frequency of Amino Acid appearance in the include/exclude metabolite lists for S. oneidensis.
While twelve amino acids were not included in the original de novo biomass proposals of BioMog (Table 1), the growth phenotypes do not preclude their addition as shown here. Their exclusion by BioMog is a result of insufficient evidence for their inclusion based soley on the growth phenotype experiments performed. Metabolites in green are those that were selected by BioMog as essential biomass components. Metabolites in blue are those that have evidence to support their inclusion but were omitted because they are redundant with other metabolites for matching experimental results (e.g., isoleucine, leucine and valine biosynthesis pathways share many of the same enzymes). Metabolites colored red have no growth phenotype evidence to support their inclusion or exclusion from the biomass.
Table 2.
BioMog Proposed De Novo Biomass for S. oneidensis using Growth Phenotype and Experimentally Measured Biomass.
Table 3.
BioMog Modified Predefined Biomass for S. oneidensis using Growth Phenotype and Experimentally Measured Biomass.