Empirical Dynamic Modeling for Accurate Prediction and Detection of State Transition in Physiological and Behavioral...

Accurate prediction of state transitions in timeseries of biological data is crucial for understanding the dynamics of physiological and behavioral responses in animal models. Traditional methods such as autoregressive integrated moving average (ARIMA), support vector regression (SVR), and convolutional long short-term memory (LSTM) networks have been widely applied but often fall short due to the inherent nonlinear and nonstationary nature of biological systems. Here we explore Empirical Dynamic Modeling (EDM) as a novel approach to dynamically predict key physiological and behavioral parameters in mice. Our results demonstrate that EDM outperforms the traditional methods in predictive accuracy and exhibits heightened sensitivity to critical physiological transitions, particularly those induced by external perturbations. Our study highlights the potential of EDM as a powerful tool for analyzing biological dynamics in preclinical research and its significant implications in identifying state transitions for human behavior and just-in-time interventions.