Inferring synaptic inputs from spikes with a conductance-based neural encoding model

Abstract

A popular approach to the study of information processing in the nervous system is to char-acterize neural responses in terms of a cascade of linear and nonlinear stages: a linear filter to describe the neurons stimulus integration properties, followed by a rectifying nonlinearity to convert filter output to spike rate. However, real neurons integrate stimuli via the modula-tion of nonlinear excitatory and inhibitory synaptic conductances. Here we introduce a bio-physically inspired point process model with conductance-based inputs. The model provides a novel interpretation of the popular Poisson generalized linear model (GLM) as a special kind of conductance-based model, where excitatory and inhibitory conductances are modulated in a \"push-pull\" manner so that total conductance remains constant. We relax this constraint to obtain a more general and flexible \"conductance-based encoding model\" (CBEM), which can exhibit stimulus-dependent fluctuations in gain and dynamics. We fit the model to spike trains of macaque retinal ganglion cells and show that, remarkably, we can accurately infer underlying inhibitory and excitatory conductances, using comparisons to intracellularly measured conductances. Using extracellular data, we corroborate the intracellular finding that synaptic excitation temporally precedes inhibition in retina. We show that the CBEM outperforms the classic GLM at predicting retinal ganglion cell responses to full-field stimuli, generalizes better across contrast levels, and captures inhibition-dependent response properties to spatially structured stimuli. The CBEM provides a powerful tool for gaining insights into the intracellular variables governing spiking, and forges an important link between extracellular characterization methods and biophysically detailed response models.