Uncovering temporal structure in hippocampal output patterns

Abstract

Place cell activity of hippocampal pyramidal cells has been described as the cognitive substrate of spatial memory. Replay is observed during hippocampal sharp-wave-ripple-associated population burst events (PBEs) and is critical for consolidation and recall-guided behaviors. PBE activity has historically been analyzed as a phenomenon subordinate to the place code. Here, we use hidden Markov models to study PBEs observed in rats during exploration of both linear mazes and open fields. We demonstrate that estimated models are consistent with a spatial map of the environment, and can even decode animals’ positions during behavior. Moreover, we demonstrate the model can be used to identify hippocampal replay without recourse to the place code, using only PBE model congruence. These results suggest that downstream regions may rely on PBEs to provide a substrate for memory. Additionally, by forming models independent of animal behavior, we lay the groundwork for studies of non-spatial memory.

Unsupervised Clusterless Decoding using a Switching Poisson Hidden Markov Model

“…With overwhelmingly short sequences, we can not expect the estimation of the transition probability matrix to be as reliable as if we had longer sequences, but we still find that the estimated transition probability matrix (Figure 7.C) is sparse, and strongly clustered around the diagonal and super-diagonal, with only a few other transitions scattered throughout. This sparse structure is suggestive of sequential hippocampal dynamics, as we would expect during running behavior (Maboudi et al, 2018).…”

“…For the unknown number of states, a more advanced Bayesian treatment may be considered, where the number of states is itself learned directly from the data such as in the hierarchical Dirichlet process hidden Markov model (see e.g., Teh et al, 2006), or a likelihood-based approach may be followed, as described by Celeux and Durand (2008). However, as far as decoding accuracy is concerned, it may be argued that the HMM approach is largely insensitive to the precise number of states (see e.g., Maboudi et al, 2018, as well as the example in this section). In effect, choosing a larger number of hidden states partitions our latent space into a finer partition, and conversely, a lower number of states specifies a coarser partition.…”

“…In particular, a rat was trained to run back-and-forth on a 100 cm long linear track for liquid reward, and we subsequently recorded neural activity while the animal was performing this task. We expected that the clusterless HMM would capture the characteristic ensemble population "place cell" activity during periods in which the animal was running, and we expected to see that latent states correspond roughly to locations along the linear track (see e.g., Linderman et al, 2016;Maboudi et al, 2018).…”

“…A complimentary paradigm to the supervised decoding approaches mentioned above, is that of unsupervised approaches to understand and interpret neural data (see e.g., Low et al, 2018;Williams et al, 2018;Maboudi et al, 2018;Mackevicius et al, 2019). These unsupervised approaches provide powerful ways of understanding the latent dynamics of neural activity, but to date they all require sorted spikes from ensembles of neurons.…”

“…In this paper, we extend one particular unsupervised approach, namely the (switching Poisson) hidden Markov model (HMM; see e.g., Linderman et al, 2016;Maboudi et al, 2018, as well as Figure 1.E) to the clusterless setting, in a new model that we call the clusterless hidden Markov model. This clusterless HMM builds on the switching Poisson framework used by Gat and Tishby (1993); Kemere et al (2008), and others, by treating the neuron identities responsible for the spikes as latent variables in the model (see Figure 2.F), and by incorporating our uncertainty about those identities in the form of parametric distributions of the waveform features associated with each hidden neuron, the mark distributions.…”