Extended Data Fig. 5: Dissociating sensory and motor coding in example V1 cells using a regression model.

(a) Each heatmap shows the firing rate over time for a subset of trials with each row representing a different trial type, and each column a different source of the firing rate. All data from MST mice. The leftmost column shows the original firing rate. The second column shows the predicted firing rate for the same trials using all predictors in the model. The remaining columns show the predicted firing rate using only a subset of the predictors. For this example neuron, the trial number explained little variability (trial number captured response drift across the session for some other neurons, not shown). Visual predictors explained an early response transient especially in Vmax trials. Auditory predictors captured an early response transient in some auditory trials (set C/D), whereas motor variables (the first 25 video PCs) captured variability across visual and auditory trial types. (b) An example neuron with strong, short-latency auditory-related firing and little motor-related activity. Same as a, but only for auditory trials and with predicted firing based on trial number and visual stimuli omitted. (c) Same as b, but for an example neuron with strong motor-related firing and little auditory-related activity. These examples (b, c) show that the trial-to-trial variability in motor activity and the variability in auditory stimulus type permit dissociation of the origin of sound-evoked activity.