Benchmarks for identification of ordinary differential equations from time series data

Area of Science:

• Systems Biology
• Computational Biology
• Mathematical Modeling

Background:

• Numerous methods exist for identifying ordinary differential equation (ODE) models from biological time series data.
• Evaluating and comparing these methods is challenging due to inconsistent problem specifications and a lack of standardized benchmark datasets.
• This hinders objective assessment of model identification algorithm performance.

Purpose of the Study:

• To establish a standardized framework for defining and evaluating ODE model identification problems.
• To create a comprehensive collection of benchmark problems for assessing computational methods in systems biology.
• To facilitate reproducible and comparable performance evaluations of different ODE identification algorithms.

Main Methods:

• Proposed a method to define ODE identification problems as optimization problems, specifying model spaces, parameter ranges, time series data, and error functions.
• Defined a standardized file format for these benchmark problems.
• Developed a collection of over 40 benchmark problems for ODE model identification in cellular systems, including simulated and real data.

Main Results:

• The proposed framework enables clear specification and comparison of ODE identification problems.
• The benchmark collection covers diverse complexities, aiding evaluation across various data qualities and system sizes.
• Our identification algorithm successfully solved all benchmark problems, often without needing supercomputing, outperforming previous approaches.

Conclusions:

• The developed benchmark collection and problem specification framework significantly improve the evaluation and comparison of ODE identification methods.
• This standardization facilitates more reliable and efficient development of computational tools for systems biology.
• The availability of these benchmarks promotes reproducible research and objective assessment of model identification algorithms.