Approximate parameter inference in systems biology using gradient matching: a comparative evaluation.

Area of Science:

• Systems Biology
• Computational Biology
• Biophysics

Background:

• Parameter inference in biological pathways modeled by ordinary differential equations (ODEs) is computationally intensive.
• Conventional numerical ODE solvers incur significant computational costs.
• Gradient matching methods offer a computationally efficient alternative by avoiding numerical integration.

Purpose of the Study:

• To present and evaluate an adaptive gradient matching approach using Gaussian processes (GPs) combined with parallel tempering for ODE parameter inference.
• To compare the proposed GP-based method against state-of-the-art techniques, including reproducing kernel Hilbert spaces (RKHS).
• To assess the robustness of the RKHS method under various experimental settings and improve its stability using cross-validation.

Main Methods:

• Adaptive gradient matching using Gaussian processes (GPs).
• Parallel tempering scheme for enhanced exploration of the parameter space.
• Reproducing kernel Hilbert spaces (RKHS) method for comparative analysis.
• Cross-validation for penalty parameter inference in the RKHS method.

Main Results:

• The proposed GP method with parallel tempering demonstrated superior performance for parameter inference when noise variance was known.
• The RKHS method showed greater robustness when the noise variance was unknown.
• Comparative analysis was conducted on two benchmark ODE systems under varying experimental conditions.

Conclusions:

• The GP-based method with parallel tempering is highly effective for ODE parameter inference with known noise variance.
• The RKHS method offers a more robust alternative for parameter inference when noise variance is unknown.
• The study highlights the trade-offs between different parameter inference techniques based on noise variance assumptions.