Title: Pairwise maximum entropy models explain correlated activity of neural populations in the inferior temporal cortex of macaque monkeys
Authors: *H. SHIOZAKI1, T. MOTONAGA2, H. TAMURA1, I. FUJITA1;
1Grad. Sch. of Frontier Biosci., 2Sch. of Engin. Sci., Osaka Univ., Toyonaka, Japan
Abstract: Correlations between the spike activities of different neurons are
ubiquitous in the cerebral cortex, meaning that understanding the neural codes
requires a description of the specific neuronal population activity. Recent
studies have shown that spatial patterns describing the activity of tens of
retinal ganglion cells and neurons in the primary visual cortex, a lower order
visual area, can be explained by a model that considers the firing rates of
individual cells and pairwise correlations between neurons without the need
to incorporate higher-order correlations. However, it remains unclear whether
pairwise interactions account for the concerted activity of neurons in higher-order
visual areas where the dendritic morphology and local anatomical connections
differ from those in early visual processing stages. To address this issue,
we simultaneously recorded the spike activity of a population of neurons in
the inferior temporal cortex of two monkeys under analgesia. We obtained data
from 9 populations consisting of 23 to 45 neurons over 15 to 60 minutes. The
probability of the number of neurons simultaneously active in the same 5-ms
time bin systematically deviated from the predictions given by a model that
assumes statistical independence between the activity of different neurons.
As an alternative, we derived a maximum entropy model that takes into account
pairwise correlations and the firing rates of individual neurons. Our resulting
model captured 94% of the departures seen in the statistical independence model.
We further tested our model by examining how well it predicts the probability
distribution of instantaneous firing patterns in a 5-ms time bin. We randomly
selected subpopulations consisting of 10 neurons and then derived the pairwise
maximum entropy model for each. The pairwise maximum entropy models predicted
86% of the departures that occurred in the statistical independence models.
From these results, we conclude that pairwise interactions account for the
correlated activity of neurons in the inferior temporal cortex, similar to
early visual areas, and speculate that synchronized firing in different visual
areas have common characteristics despite their heterogeneous anatomical structures.