Evaluating the limitations of Bayesian metabolic control analysis.

Area of Science:

• Systems Biology
• Metabolic Engineering
• Computational Biology

Background:

• Bayesian Metabolic Control Analysis (BMCA) infers metabolic control coefficients using Bayesian inference and lin-log rate laws.
• These coefficients are crucial for understanding how enzyme activity changes impact metabolic network steady states.
• The predictive accuracy and limitations of BMCA, especially in data-limited scenarios, require thorough investigation.

Purpose of the Study:

• To systematically evaluate BMCA's performance in inferring elasticity values, flux control coefficients (FCC), and concentration control coefficients (CCC).
• To assess the impact of varying data availability, including flux, enzyme concentration, and external metabolite concentration data, on BMCA's predictive accuracy.
• To compare the performance of ADVI and HMC inference engines and identify limitations in elasticity and allosteric interaction recovery.

Main Methods:

• Utilized three synthetic metabolic network models to simulate various data availability conditions.
• Performed systematic evaluations of BMCA's ability to infer elasticity values, FCC, and CCC.
• Compared inference accuracy using ADVI and HMC, focusing on elasticity magnitude and allosteric interaction recovery.

Main Results:

• BMCA predictions are highly sensitive to the inclusion of flux and enzyme concentration data; their omission leads to significant inaccuracies.
• External metabolite concentrations had minimal impact, and their exclusion sometimes improved predictions.
• Both ADVI and HMC underestimated large-magnitude elasticities (|elasticity| >= 1.5), with ADVI showing higher variance under strong up-regulation.
• ADVI failed to accurately infer strong allosteric interactions.
• BMCA partially recovered rankings of high FCC values, but absolute value estimates were constrained by priors and data limitations.

Conclusions:

• BMCA's predictive accuracy is strongly contingent on the availability and quality of flux and enzyme concentration data.
• Current inference engines struggle to accurately recover large-magnitude elasticities and complex allosteric interactions.
• BMCA offers value in ranking key control coefficients but requires methodological improvements for precise quantitative predictions in metabolic engineering.