Systematic reduction of the dimensionality of natural scenes allows accurate predictions of retinal ganglion cell spike outputs

Abstract The mammalian retina engages a broad array of linear and nonlinear circuit mechanisms to convert natural scenes into retinal ganglion cell (RGC) spike outputs. Although many individual integration mechanisms are well understood, predictive models of natural scene encoding perform poorly, likely due to interactions among the active mechanisms. Here, we identified spatial integration mechanisms relevant for natural scene encoding and used computational modeling to predict spike outputs in primate parasol RGCs. Our approach was to simplify the structure of natural scenes and empirically ask whether these changes were reflected in RGC spike responses. We observed that reducing natural movies to 16 linearly integrated regions described ∼80% of the structure of parasol RGC spike responses. We then used simplified stimuli to create high-dimensional metamers that recapitulated the spike response of full-field naturalistic movies. Finally, we identified the retinal computations that convert natural images in 16-dimensional space into 1-dimensional spike outputs.