Latent variable models for hippocampal sequence analysis

Place cell activity of hippocampal pyramidal cells has been described as the cognitive substrate of spatial memory. Indeed, the activity of ensembles of neurons within the hippocampus is thought to enable memory formation, storage, recall, and even decision making. Replay is observed during hippocampal sharp-wave-ripple-associated population burst events (PBEs) and is critical for consolidation and recall-guided behaviors. Notably, these PBEs occur during times of inactivity, so that their representations cannot easily be matched with observable animal behavior. In my thesis, I present an approach to uncover temporal structure within hippocampal output patterns during PBEs. More specifically, I use hidden Markov models (HMMs) to study PBEs observed in rats during exploration of both linear tracks and open fields, and I demonstrate that estimated models are consistent with a spatial map of the environment. Moreover, I demonstrate how the model can be used to identify hippocampal replay without recourse to the place code. These results suggest that downstream regions may rely on PBEs to provide a substrate for memory. Moreover, by forming models independent of animal behavior, I lay the groundwork for studies of non-spatial memory. Next, I present a new model, the "clusterless" switching Poisson hidden Markov model, which extends my work on HMMs of PBEs to the case where we only have multiunit (unsorted) spikes. Indeed, spike sorting is challenging, time-consuming, often subjective (not reproducible), and throws away potentially valuable information from unsorted spikes, as well as our certainty about the cluster assignments. It has previously been shown that we can often do just as well, or in some cases even better, if we forego the spike sorting process altogether, and work directly with the unsorted data. Consequently, my clusterless HMM will enable us to combine the benefits of unsupervised learning for internally generated neural activity, with the benefits of clusterless approaches (more data leading to higher fidelity, especially at fine temporal scales, and additional probabilistic / soft information to exploit). I demonstrate the model's ability to recover model parameters for simulated data, and show that it is able to learn a spatially-consistent representation of the environment from real experimental data.

Name thesis:degree_name

Doctor of Philosophy

Level thesis:degree_level

Doctoral

Discipline thesis:degree_discipline

Engineering

Grantor

Rice University

Year dc:date.issued

2019

Author dc:creator

• Ackermann, Etienne Rudolph

Advisor dc:contributor.advisor

• Kemere, Caleb

• hippocampal replay
• hidden Markov model

Statement dc:rights

• Copyright is held by the author, unless otherwise indicated. Permission to reuse, publish, or reproduce the work beyond the bounds of fair use or other exemptions to copyright law must be obtained from the copyright holder.

Language dc:language.iso

eng

Handle dc:identifier.uri

https://hdl.handle.net/1911/106148

OAI identifier oai:identifier

oai:repository.rice.edu:1911/106148