Inferring the role of inhibition in auditory processing of complex natural stimuli

Abstract

Intracellular studies have revealed the importance of cotuned excitatory and inhibitory inputs to neurons in auditory cortex, but typical spectrotemporal receptive field models of neuronal processing cannot account for this overlapping tuning. Here, we apply a new nonlinear modeling framework to extracellular data recorded from primary auditory cortex (A1) that enables us to explore how the interplay of excitation and inhibition contributes to the processing of complex natural sounds. The resulting description produces more accurate predictions of observed spike trains than the linear spectrotemporal model, and the properties of excitation and inhibition inferred by the model are furthermore consistent with previous intracellular observations. It can also describe several nonlinear properties of A1 that are not captured by linear models, including intensity tuning and selectivity to sound onsets and offsets. These results thus offer a broader picture of the computational role of excitation and inhibition in A1 and support the hypothesis that their interactions play an important role in the processing of natural auditory stimuli.

Network Receptive Field Modeling Reveals Extensive Integration and Multi-feature Selectivity in Auditory Cortical Neurons

“…Another extension to GLMs, a generalized nonlinear model (GNM), does, however, employ input units with monotonically-increasing nonlinearities, and unlike multi-neuron GLMs or GQMs, GNMs have been applied to auditory neurons by Schinkel-Bielefeld and colleagues [24]. Their GNM comprises a very simple feedforward network based on the weighted sum of an excitatory and an inhibitory unit, along with a post-spike filter.…”

“…Unlike the above two GNM [24,53] approaches, this model does not have hidden units with sigmoidal nonlinearities, but finds multiple quadratic features. The MNE predicted neural responses better than a linear model, although still poorly, with an average CC raw of 0.24, and it was not determined whether it could out-predict an LN model.…”

“…The NRF model is composed of a hierarchical feedforward network of 20 LN units, embodying the fact that cortical neurons integrate the output of many lower order neurons. Although our choice of 20 possible feed-forward connections does not reflect the full range of converging inputs that cortical neurons receive, this approach stands in contrast to the above mentioned recent models of A1 responses [20–24], which use only one, or in some cases two, STRF-like units. In fact, we show here that up to seven effective units are required to model a cortical receptive field.…”

Incorporating Midbrain Adaptation to Mean Sound Level Improves Models of Auditory Cortical Processing

“…Although STRF-based models have their limitations, they remain valuable and provide simple models for describing the behavior of sensory neurons. Many of the more sophisticated models (Sharpee et al, 2004Ahrens et al, 2008;Atencio et al, 2009;Calabrese et al, 2011;Schinkel-Bielefeld et al, 2012) require many more parameters and may also be difficult to interpret biologically. Thus, a key challenge is to extend STRF models so that they more accurately describe neuronal behavior, while remaining simple and biologically relevant.…”

Models of Neuronal Stimulus-Response Functions: Elaboration, Estimation, and Evaluation

“…Such a model has also been called a generalized nonlinear model (GNM) (Butts et al, 2007, 2011; Schinkel-Bielefeld et al, 2012), or nonlinear input model (NIM) (McFarland et al, 2013) and model parameters may be estimated by maximizing the spike-train likelihood of an inhomogeneous Poisson model with rate given by Equation (17)—often using a process of alternation similar to that described above. Vintch et al (2015) combined an LNLN model with basis-function expansion of the nonlinearity (Equation 15) to yield a model for visual responses of the form…”