Abstract
We evaluated how neuronal interactivity changes with visual stimulus complexity in mouse subcortical and cortical visual areas. To do this, we leveraged Integrated Information Theory (IIT4.0; Mayner et al., 2023) to derive a measure for neuronal interactivity termed PHI hat (ϕ̂). This novel measure, which slightly deviates from IIT’s measure PHI, allows us to go beyond pairwise analysis and quantify higher-order causal interactions in population spiking data to better understand neural complexity. We used an open-source dataset collected by the Allen Institute (Siegle et al., 2022). From this dataset, we used 52 Neuropixel recordings spanning six visual areas. We filtered the local field potential (LFP) data for gamma-range activity and binarized our data around the median. We then created a state-by-state transition matrix that quantified the probability of transitioning from a given state at time point t to any other possible state at time t + 1 (Leung et al., 2021). Using the open-source PyPhi toolbox (Mayner et al., 2018), we calculated 4 values of ϕ̂ for each presentation in a system of three randomly selected channels 3 times per probe. We found that ϕ̂ increases with visual stimulus complexity, from static gratings to natural movies across all visual areas (p < .01, ANOVA). However, we found no statistically significant difference in ϕ̂ values between brain regions (p > 0.227, ANOVA). Even early visual areas such as the LGd and higher areas beyond V1 exhibited similar ϕ̂ values for each type of stimulus. Both the increase in neuronal interactions as a function of stimulus complexity and the lack of clear contrast between cortical areas match a previous finding on neural differentiation (Mayner et al., 2022). Taken together, these findings suggest that ϕ̂ is a promising measure of complexity.