A comparison of Monte Carlo sampling methods for metabolic network models

Area of Science:

• Systems Biology
• Metabolic Engineering
• Computational Biology

Background:

• Constraint-based modeling (CBM) is crucial for analyzing metabolic networks under steady-state assumptions.
• Deterministic CBM requires exact flux fulfillment, neglecting experimental noise.
• Stochastic CBM relaxes steady-state constraints and incorporates noise for a more realistic representation.

Purpose of the Study:

• To provide an overview of deterministic and stochastic formulations in metabolic network analysis.
• To evaluate the performance of various Monte Carlo sampling methods for flux analysis.
• To compare the efficiency, consistency, and convergence of sampling algorithms.

Main Methods:

• Overview of deterministic and stochastic constraint-based modeling approaches.
• Application and evaluation of four Monte Carlo sampling algorithms: ACHR, OPTGP, CHRR, and Gibbs sampler.
• Analysis of sampling performance across ten diverse metabolic networks.

Main Results:

• Coordinate hit-and-run with rounding (CHRR) demonstrated superior performance and guaranteed convergence for deterministic formulations.
• ACHR showed high consistency with CHRR for genome-scale models.
• The Gibbs sampler was the only feasible method for genome-scale stochastic analysis but was less efficient than deterministic samplers.

Conclusions:

• CHRR is the recommended algorithm for deterministic metabolic flux analysis due to its efficiency and convergence properties.
• For genome-scale stochastic analysis, the Gibbs sampler is currently the most suitable, despite its lower efficiency.
• Further research may focus on improving the efficiency of stochastic sampling methods for large-scale metabolic networks.