Time localized trigger for phase synchronization transition in complex neuronal networks

Understanding how the brain exploits dynamical interactions to support intelligent behavior remains a central challenge in neuroscience. Neuronal synchronization plays a key role in coordinating information processing, but its intermittent nature involves many interacting factors, making it difficult to anticipate changes in network state. In this work, we investigate the dynamical signatures that precede transitions in synchronization within simulated neuronal networks. We show that preferential time windows emerge before these transitions and can be used to anticipate upcoming shifts in the synchronization regime. Using a Random Forest classifier, we predict these changes with over 80% accuracy in a network configuration where transition times are strongly influenced by the neurons’ natural frequency distribution. We also identify which neurons contribute most to triggering the transition, providing insight into the mechanisms of synchronization loss and recovery. These predictive signatures emerge before changes appear in conventional synchronization measures, indicating that machine-learning approaches can uncover precursors of dynamical transitions in neuronal systems. These results highlight the potential of data-driven methods to anticipate synchronization shifts and to reveal the dynamical mechanisms of intermittent activity in complex neuronal networks.