Constrained Fourth Order Latent Differential Equation Reduces Parameter Estimation Bias for Damped Linear Oscillator...

Area of Science:

• Mathematical modeling
• Systems biology
• Time series analysis

Background:

• Second order linear differential equations model regulatory systems, exhibiting equilibrium return or oscillations.
• Latent Differential Equations (LDE) estimate parameters from time series data.
• LDE parameter estimation can be biased by suboptimal time delay embedding dimensions and convolution kernel widths.

Purpose of the Study:

• To investigate bias reduction in parameter estimation for second order systems using a novel modeling approach.
• To evaluate the performance of a constrained fourth order Latent Differential Equation (FOLDE) model.

Main Methods:

• Simulation study using time series data.
• Application of a constrained fourth order Latent Differential Equation (FOLDE) model.
• Analysis of parameter bias in relation to convolution kernel width and oscillation period.

Main Results:

• The constrained FOLDE model significantly reduces parameter bias in second order systems.
• Bias elimination is nearly complete when the convolution kernel width is less than two-thirds of the oscillation period.
• The FOLDE model, with two additional degrees of freedom, offers substantial improvement in model fit compared to standard LDE.

Conclusions:

• The constrained FOLDE model provides a more accurate parameter estimation for second order systems, especially in the presence of oscillations.
• Careful selection of the convolution kernel width is crucial for minimizing bias in LDE-based modeling.
• FOLDE offers enhanced accuracy and model fit for analyzing dynamic systems.