Short-Range Temporal Interactions in Sleep; Hippocampal Spike Avalanches Support a Large Milieu of Sequential Activity Including Replay
Fig 6
The historical context of firing modulates the probability of firing for the cells in the ensemble. The strength of this modulation can be measured as a ‘conditional log-likelihood ratio’ comparing the fitted model to the avalanche independent model. These ratios can be organized into rectangular matrices, denoted by R, with contexts in rows and symbols in columns (see Materials and Methods). (A) Heatmap of the R matrix of a representative example (Rat 1, session 2) has clusters of contexts that similarly modulate the ensemble (black lines). The color scale represents the log-likelihood ratio between the fitted model and the AIM; red indicated up-regulation, blue indicates down-regulation. We have set the color scale between -1 (2-fold down-regulation) and +1 (2-fold up-regulation) to clarify the clustering structure, but note that these values have greater variation (Fig 5B). In Rat 1, session 2, clusters 1 and 2 strongly differentially regulate cells 1,3 7, and 9. This indicates that the cells represented in the contexts from different clusters form cell assemblies of coactivating cells. Note however that there is wide variation within each cluster indicating that each context indeed uniquely modulates the ensemble. (B) To visualize the clustering, we computed the cosine similarity matrices between pairs of contexts in the R matrices. These similarity matrices encode how similarly two contexts modulate the ensemble. A value of 1 indicates that the contexts make identical predictions about the upcoming neuron to fire, while a value of 0 indicates that the contexts are extremely dissimilar. The similarity matrices have clear block structure indicating that the clustering in (A) is a strong feature of the R matrices (labels of four prominent clusters shown at the top of the similarity matrix).