Previously, we have developed a Bayesian approach to infer rat hippocampal population codes during spatial navigation. The population neuronal spike trains are modeled by a hidden Markov model (HMM), which is driven by a latent Markov chain. To tackle the model selection problem for the latent state, we further leverage a nonparametric Bayesian model. Specifically, we develop a hierarchical Dirichlet process-HMM (HDP-HMM) and two Bayesian inference methods, one based on Markov chain Monte Carlo (MCMC) and other based on variational Bayes (VB). The effectiveness of Bayesian approaches is demonstrated on recordings from freely-behaving rats navigating in an open field environment. Based on these empirical studies, we further apply our analysis tools to analyze rat hippocampal ensemble spike data in a more challenging setting. In contrast to wake, during sleep there is a complete absence of animal behavior, and the ensemble spike activity is fragmental in time. How can we pursue statistical analysis towards sleep-associated hippocampal ensemble spike activity? To address the question, we construct synthetic sleep-alike hippocampal spike data. Using Bayesian population decoding analysis, we systematically investigate the impacts of (i) individual epoch length, (ii) number of neurons, and (iii) temporal scale. Our analyses suggest that under sleep-alike conditions (short epochs, low spiking, compressed timescale), the representation capacity and recovery accuracy degrade; however, it is still possible to recover the spatial topology of the 2D environment. These results provide further insight for sleep-associated ensemble spike data analysis.
展开▼
