Skip to main content
. 2020 Jan 22;10(1):109–119.e3. doi: 10.1016/j.cels.2019.11.006

Figure 2.

Figure 2

Ensemble Generation Process and Subsampling of Ensemble Content and Simulations to Demonstrate Adequate Sampling of the Solution Space

(A) Algorithm for generating a single ensemble member. Given a set of confirmed experimental growth conditions, reactions are added to fill gaps in the GEM required for in silico growth in the first condition (see STAR Methods, Algorithm 1). The reactions are added, then the process is performed iteratively for the remaining growth conditions until growth is possible in all conditions.

(B) Process for producing an ensemble of GEMs. The algorithm in (A) is performed to generate a single ensemble member, then the order of the media conditions is shuffled, and the process is repeated to generate another ensemble member. This process is repeated 1,000 times, generating approximately 1,000 ensemble members (for some species, duplicate solutions occur and these members are removed. All 29 species had 970–1,000 ensemble members).

(C) The variable reaction content in ensemble members as a function of increasing ensemble size. A variable reaction is any reaction that is variably included across any member of the subsampled group (e.g., it is absent from some members and present in other members but neither entirely present in nor entirely absent from the ensemble). For each species, the mean number of variable reactions in subsamples of GEMs is shown by the solid line, with the standard deviation shown as light fill of the same color above and below the mean. Subsampling was performed with 1,000 draws per subsample size. Ensembles were sampled at intervals of 20 members, e.g., 20, 40, 60, and so on until reaching the size of the entire ensemble.

(D) Variability in gene essentiality simulations within subsamples of ensemble members. Using the same subsampling procedure as in (C), the number of genes with at least one GEM in the subsample with a simulation outcome different than the rest (e.g., non-consensus) was determined. The mean for each subsample size is shown by the solid line, with the standard deviation shown as light fill of the same color above and below the mean.