Scalable Bayesian Stochastic Process Models for Neural Data Analysis

总结

Summary The overarching goal of this proposal is to integrate the development of a novel class of statistical methods with unique electrophysiological experiments in rats to address fundamental and unresolved questions about hippocampal function and, in subsequent studies, to provide unprecedented insight into the neural mechanisms underlying memory impairments. The development of novel statistical tools for the analysis of neural data is key to advance our understanding of fundamental memory mechanisms and of memory disorders. However, many existing statistical methods are not capable of handling such data-intensive problems in terms of theoretical foundation, computational complexity, and scalability. To address this issue, we will design a robust framework for analyzing neural data using flexible multivariate Gaussian process (GP) models (Aim 1). This novel framework will allow the integration of multiple data modalities, in particular multi- neuronal spike trains and multi-node local field potentials (LFP), while identifying their joint low-dimensional representations and underlying structures. To make our approach practical for big data analysis, we will develop computationally efficient algorithms for fast, yet accurate statistical inference (Aim 2). Our proposed approach is based on a novel combination of fast variational approximation methods and computationally efficient Markov Chain Monte Carlo algorithms. We will apply our analytical methods to unique electrophysiological datasets collected as part of a research program aimed at elucidating the fundamental neural mechanisms underlying the memory for sequences of events, a defining feature of episodic memory (Aim 3). In these datasets, we use high-density electrophysiological techniques to record neural activity in hippocampal region CA1 (spikes and LFP) as rats perform an odor sequence memory task. Importantly, this nonspatial approach allows us to determine whether spatial coding properties (thought to be fundamental to hippocampal memory function) extend to the nonspatial domain, including sequence reactivation (reactivation of previously traversed or upcoming sequences of locations) and phase precession (spikes occurring at progressively earlier phases of the theta cycle during traversal of the cell's place field). Combining these novel analytical tools with the sophisticated behavioral, electrophysiological, and DREADDs inactivation approaches proposed here will provide us with an unparalleled opportunity to address fundamental questions about hippocampal function.