Drift-diffusion dynamics of the hippocampal replay

Replay activities in the hippocampus and other brain regions during sharp-wave ripples (SWRs) are thought to play important roles in learning, memory, and planning. Surprisingly, the question of how to characterize the dynamical structure of replay remains controversial. Standard methods rely on restrictive assumptions for detecting replay events with high sequentiality, and they have substantial drawbacks. To fill this gap, we develop a flexible and highly interpretable computational framework based on state-space modeling to understand the dynamics of replays. This framework is motivated by the basic idea that sequential structures can be modeled using drift-diffusion dynamics. The two parameters (i.e., drift & diffusion parameters) can be directly mapped to the speed and variability of a replay event, respectively. To capture the potentially rich neural dynamics during SWRs, our model allows the switching between several well-motivated types of dynamics. The resulting framework enables precise and unambiguous interpretations of the dynamics of SWR events. Applications of our method to population recordings from the rat hippocampus lead to insights into a number of important open questions, including: (i) whether the speed of most replay events is comparable to real-world running; (ii) whether replays follow random walk; (iii) whether “preplay” events exist. More broadly, accurate characterizations of replay dynamics may lead to a better understanding of the role of replay under normal and abnormal conditions.