Composing problem solvers for simulation experimentation: a case study on steady state estimation.

Area of Science:

• Systems Biology
• Computational Biology
• Scientific Simulation

Background:

• Simulation experiments require diverse sub-tasks like parameter optimization and data analysis.
• Algorithm performance varies depending on the specific problem, necessitating careful selection by researchers.
• Steady state estimation in systems biology exemplifies this challenge, with multiple estimators having distinct advantages and disadvantages.

Purpose of the Study:

• To propose a general scheme for aggregating algorithms into synthetic problem solvers.
• To improve overall performance in simulation tasks by exploiting algorithm differences.
• To support experimenters in choosing appropriate algorithms by providing a robust aggregation framework.

Main Methods:

• Developed a general scheme to aggregate algorithms into synthetic problem solvers.
• Incorporated various aggregation mechanisms, including ensemble learning and portfolio selection.
• Extended the plugin system of the open-source modeling and simulation framework James II.

Main Results:

• Demonstrated improved overall performance by exploiting differences between algorithms.
• Enabled automatic configuration of solvers using training data.
• Successfully applied the approach to steady state estimation for cell-biological models.

Conclusions:

• The proposed synthetic problem solver scheme effectively enhances performance in simulation tasks.
• This approach provides a flexible and data-driven method for algorithm aggregation.
• The framework offers practical benefits for researchers in systems biology and other simulation-intensive fields.