Nonlinearity identified by neural network models in Pco2 control system in humans

Area of Science:

• Physiology
• Computational Neuroscience
• Systems Biology

Background:

• The human respiratory system's regulation of carbon dioxide (CO2) is crucial for homeostasis.
• Understanding the control system's linearity or nonlinearity is essential for accurate physiological modeling.

Purpose of the Study:

• To evaluate the degree of nonlinearity in the human PCO2 control system.
• To compare the effectiveness of different modeling approaches, including autoregressive moving average (ARMA) and neural networks.

Main Methods:

• Applied an autoregressive moving average (ARMA) model and linear/nonlinear neural networks to model the PCO2 control system.
• Utilized a difference of residuals to quantify the degree of nonlinearity.
• Compared Jordan, Elman, and fully interconnected neural network architectures.

Main Results:

• A linear Jordan-type neural network failed to accurately approximate respiratory data.
• The ARMA model and nonlinear neural networks (Elman, fully interconnected) were used for nonlinearity evaluation.
• The linear assumption for the PCO2 control system was found to be invalid for 3 out of 7 subjects.
• Strong nonlinear dynamics were specifically observed in 2 subjects.

Conclusions:

• The human PCO2 control system demonstrates significant nonlinearity in a subset of individuals.
• Linear modeling approaches are inadequate for capturing the complex dynamics of respiratory control in these subjects.
• Nonlinear modeling techniques, particularly advanced neural networks, are necessary for accurate representation.