Reconciling high-throughput gene essentiality data with metabolic network reconstructions

Abstract

The identification of genes essential for bacterial growth and survival represents a promising strategy for the discovery of antimicrobial targets. Essential genes can be identified on a genome-scale using transposon mutagenesis approaches; however, variability between screens and challenges with interpretation of essentiality data hinder the identification of both condition-independent and condition-dependent essential genes. To illustrate the scope of these challenges, we perform a large-scale comparison of multiple published Pseudomonas aeruginosa gene essentiality datasets, revealing substantial differences between the screens. We then contextualize essentiality using genome-scale metabolic network reconstructions and demonstrate the utility of this approach in providing functional explanations for essentiality and reconciling differences between screens. Genome-scale metabolic network reconstructions also enable a high-throughput, quantitative analysis to assess the impact of media conditions on the identification of condition-independent essential genes. Our computational model-driven analysis provides mechanistic insight into essentiality and contributes novel insights for design of future gene essentiality screens and the identification of core metabolic processes.

Guiding the Refinement of Biochemical Knowledgebases with Ensembles of Metabolic Networks and Machine Learning

“…For both species, ensemble members have variable precision and recall, and simulation cluster membership is associated with a difference in both precision and recall (p < 0.0001, Mann-Whitney U-test with false discovery rate control via Benjamini Hochberg procedure). We note that the poor precision and recall for all ensemble members is consistent with the performance of other GEMs in predicting gene essentiality, especially when comparing to in vitro essentiality datasets that suffer from technical noise and variability (Blazier and Papin, 2019). There are biologically meaningful differences in the predictions generated by each cluster.…”

Guiding the Refinement of Biochemical Knowledgebases with Ensembles of Metabolic Networks and Machine Learning

“…For both species, ensemble members have variable precision and recall, and simulation cluster membership is associated with a difference in both precision and recall ( p < 0.0001, MannWhitney U test with false discovery rate control via Benjamini Hochberg procedure). We note that the poor precision and recall for all ensemble members is consistent with the performance of other GEMs in predicting gene essentiality, especially when comparing to in vitro essentiality datasets that suffer from technical noise and variability (Blazier and Papin, 2019) . The critical aspect of these results is that there are biologically meaningful differences in the predictions generated by each cluster.…”

An updated genome-scale metabolic network reconstruction ofPseudomonas aeruginosaPA14 to characterize mucin-driven shifts in bacterial metabolism

“…Three genes labeled as “SPONTANEOUS,” “unassigned,” and “Unassigned’’ were removed from the reconstruction given that these labels did not correspond to genes belonging to P. aeruginosa . Gene essentiality data was not used for curation of the metabolic network given the variability in gene essentiality screens and the resultant challenges with data interpretation 32 . Instead, model predictions were compared to gene essentiality data as one facet of validation.…”