Mechanistic Hierarchical Gaussian Processes.

Area of Science:

• Statistics
• Biophysics
• Computational Biology

Background:

• Functional data analysis (FDA) often uses flexible black-box models.
• Existing FDA methods struggle to incorporate mechanistic information, such as differential equations.
• Muscle activation studies require methods that can integrate physiological understanding.

Purpose of the Study:

• To develop a novel nonparametric Bayesian approach for functional data analysis.
• To incorporate mechanistic information, specifically differential equations, into FDA.
• To analyze muscle activation data by favoring curves consistent with physiological models.

Main Methods:

• Defined a new class of hierarchical Gaussian processes.
• Developed a Gibbs sampler for posterior distribution sampling.
• Applied the method to functional data from rat muscle activation studies.

Main Results:

• The proposed hierarchical Gaussian processes favor curves consistent with differential equations.
• Successfully applied the method to analyze muscle force data from rats.
• Demonstrated the utility of incorporating mechanistic information into FDA.

Conclusions:

• The developed nonparametric Bayesian approach effectively integrates mechanistic information into FDA.
• This method offers a powerful alternative to black-box approaches for analyzing complex biological data.
• The approach has broad applicability beyond muscle physiology for functional data analysis.