Transfer entropy estimation and directional coupling change detection in biomedical time series

Area of Science:

• Biomedical Engineering
• Non-linear Time Series Analysis
• Systems Physiology

Background:

• Directional coupling detection in non-linear biomedical time series is crucial, particularly for the respiratory chemoreflex system.
• Transfer entropy is a valuable tool, but its performance with small sample sizes and outliers in biomedical data is understudied.

Purpose of the Study:

• To compare the performance of different transfer entropy estimation methods in detecting changes in directional coupling within biomedical time series.
• To evaluate methods under conditions typical of biomedical applications: small sample sizes and the presence of outliers.

Main Methods:

• Compared fixed-binning with ranking, kernel density estimation (KDE), and Darbellay-Vajda (D-V) adaptive partitioning using simulated and lamb respiratory time series.
• Simulated data varied sample size (50-200) and coupling strength, incorporating outliers via Laplace distribution.
• Analyzed the influence of O2 and CO2 on ventilation (V˙E) before and after domperidone administration in lambs.

Main Results:

• Kernel density estimation (KDE) detected increased coupling strength at the lowest signal-to-noise ratio (SNR) in simulations.
• Darbellay-Vajda (D-V) partitioning showed the strongest increase in transfer entropy for PO2 → V˙E post-domperidone in lambs.
• D-V partitioning uniquely detected increased transfer entropy for PCO2 → V˙E, aligning with experimental observations.

Conclusions:

• Transfer entropy can reliably detect directional coupling changes in non-linear biomedical time series, even with limited data and outliers.
• Fixed-binning methods are insufficient; both KDE and D-V partitioning are viable, with D-V offering advantages in computational efficiency and parameter selection.
• This study provides guidance for selecting appropriate transfer entropy estimation methods in biomedical research.