Model-based transfer entropy analysis of brain-body interactions with penalized regression techniques

The human body can be seen as a functional network depicting the dynamical interactions between different organ systems. This exchange of information is often evaluated with information-theoretic approaches which comprise the use of vector autoregressive (VAR) and state space (SS) models, normally identified with the Ordinary Least Squares (OLS). However, the number of time series to be included in the model is strictly related to the length of data recorded thus limiting the use of the classical approach. In this work, a new method based on penalized regressions, the so-called LASSO, was compared with OLS on physiological time-series extracted from 18 subjects during different stress conditions. Results show similarities between the brain-body interactions estimated by both methodologies, highlighting a greater intepretability of patterns estimated with LASSO especially in the subnetwork of brain-brain interactions.