Using Neural Networks with Structured Random Connectivity for Modeling Population Responses in Mouse Auditory Cortex

Understanding how circuit connectivity shapes the structure of cortical population activity is central to neural computation. We analyze population calcium activity in mice auditory cortex subject to optogenetic and auditory stimuli with different intensities collected in [1]. We fit a minimal four-cell-type rate based model of this neural network and show that it can capture mean responses. This model, however, fails to explain the variability of population responses. We therefore test whether a “random with constraints” model can explain this data. In this model, constraints enforce the known existence or absence of interactions between neural types, but connectivity matrices are otherwise random. We compare to an unconstrained random model and to the deterministic baseline. Preliminary results indicate that the random with constraint models can quantitatively fit both the means and variances of the experimental data.

Reference

[1] Tobin M, Sheth J, Wood KC, Michel EK, Geffen MN. Distinct Inhibitory Neurons Differently Shape Neuronal Codes for Sound Intensity in the Auditory Cortex. J Neurosci. 2025;45(2):e1502232024. doi: 10.1523/JNEUROSCI.1502-23.2024. PMID: 39516042.