Scalable Tools for Analyzing Steady-State Microbial Communities Using Standardized Genome-Scale Metabolic Models

Extending genome-scale metabolic modeling to microbial communities is a promising way to unravel community interactions. We present three tools that significantly improve the accuracy and feasibility of simulating community metabolism. First, little emphasis has been placed on the need to impose a time-averaged constant growth rate across all community members to ensure co-existence and stability. Without this constraint, the faster growing organism will ultimately displace other microbes. We introduced the SteadyCom framework for predicting community metabolism consistent with the steady-state requirement. SteadyCom is scalable to a large number of organisms and compatible with flux variability analysis (FVA). In contrast to flux balance analysis, SteadyCom can predict an abundance profile with good agreement to experimental gut microbiota.

Second, during simulating community metabolism, an unstandardized biomass reaction of any organisms that produces biomass with a molecular weight different from the standard 1 g/mmol introduces a systematic error. We developed a computational procedure for checking the biomass weight so as to eliminate the error. 42 out of 64 examined models show >5% discrepancies in biomass weights. We demonstrated that biomass weight discrepancies could cause significant errors in the predicted community composition that are disproportionately larger than the biomass weight discrepancies. Microbes with underestimated biomass weights are overpredicted whereas microbes with overestimated biomass weights are underpredicted.

Third, to effectively perform FVA in the absence of thermodynamically infeasible cycles in community models, we devised a method termed localized loop-less FVA (lll-FVA) with superior computational performance. We identified the fewest constraints required under the loop-less requirement and put forth the concept of localized loop-less constraints. The computational time is reduced by 10-150 times compared to the original loop-less constraints and by 4-20 times compared to the currently fastest method Fast-SNP. lll-FVA offers a scalable strategy for loopless flux calculations for large multi-compartment/multi-organism models (e.g., >10⁴ reactions).