Parameter inference for enzyme and temperature constrained genome-scale models

Area of Science:

• Computational biology
• Systems biology
• Metabolic modeling

Background:

• Organismal metabolism is temperature-dependent, necessitating predictive models.
• Enzyme and temperature constrained genome-scale models (etcGEMs) link enzyme thermodynamics to metabolic networks.
• Constraint-based metabolic modeling is expanded by etcGEMs.

Purpose of the Study:

• To address the instability and multimodality issues in Bayesian parameter inference for etcGEMs.
• To develop a robust method for estimating temperature dependence in metabolic networks.
• To improve computational efficiency for metabolic model parameterization.

Main Methods:

• Developed an evolutionary algorithm to handle multimodal parameter spaces in etcGEMs.
• Quantified phenotypic consequences of diverse parameter solutions on key metabolic reactions.
• Optimized software for faster parameter set evaluations, reducing runtime by 8.5x.

Main Results:

• The Bayesian inference method for etcGEMs is unstable and fails with multimodal distributions.
• The evolutionary algorithm successfully explores multimodal parameter spaces, yielding diverse solutions.
• Significant phenotypic variation was observed in key metabolic reactions, indicating model under-determination.
• Computational efficiency of parameter evaluation was substantially improved.

Conclusions:

• The Bayesian approach for etcGEM parameterization is inadequate due to multimodality.
• Evolutionary algorithms offer a viable solution for parameter inference in complex metabolic models.
• Current experimental data is insufficient to fully constrain etcGEM predictions, highlighting the need for more data.
• Software improvements enable faster and more resource-efficient metabolic model analysis.