Adaptive hybrid simulations for multiscale stochastic reaction networks

Area of Science:

• Computational Biology
• Biophysics
• Chemical Kinetics

Background:

• Stochastic Reaction Networks (SRNs) are modeled by the Chemical Master Equation (CME).
• Estimating CME solutions often relies on computationally intensive methods like the Stochastic Simulation Algorithm (SSA).
• Existing hybrid methods require manual partitioning of species and reactions, limiting their applicability.

Purpose of the Study:

• To develop an automated hybrid method for approximating CME solutions.
• To overcome the computational burden of traditional SSA for SRNs.
• To adaptively partition dynamics and apply quasi-stationary approximations without user intervention.

Main Methods:

• Developed a hybrid simulation method that automatically partitions reactions and species into discrete and continuous sets.
• Incorporated adaptive quasi-stationary assumptions for fast subnetworks.
• The method dynamically adjusts partitions based on changing timescales and species copy numbers.

Main Results:

• Achieved significant computational speedups compared to exact SSA.
• Demonstrated highly accurate approximations of probability distributions for systems biology examples.
• Showcased particular efficiency for systems exhibiting oscillatory dynamics.

Conclusions:

• The proposed automated hybrid method offers a computationally efficient and accurate alternative for solving the CME.
• Adaptive partitioning and quasi-stationary approximations enhance performance for complex and dynamic SRNs.
• This method is particularly beneficial for systems biology models with time-varying dynamics.