Fig. 4: Categorical representations predict the mice’s behavioral performance.

a Orientation representational space is divided into Go (1°–60°, blue) and NoGo (61°–120°, red) domains. Go evidence is defined as the ratio of neuronal activity in the Go domain (area under the activation profile) to the total activity in both domains. b Mouse licking response probability as a function of Go evidence measured during the 200 ms before the response window onset across all sessions from two datasets. In the second dataset (red), unimodal (empty symbols) and audiovisual (solid symbols) blocks are distinguished. The black line represents a linear fit (r² = 0.87; p < 0.001). c Pearson correlation coefficients (r²) over time during visual stimulus presentation. Black: Go evidence; blue: Go domain activity; red: NoGo domain activity. Shaded areas indicate the 95% confidence interval; * indicates significant differences between correlation coefficients for Go evidence and both Go and NoGo activity. d. Average neuronal activity in orientation space evoked by a 65° NoGo stimulus (Go/NoGo angle 20°; visual-only dataset) during false alarm (FA) and correct rejection (CR) trials. e Average activity profiles for FA and CR trials over 0–2 s stimulus presentation. Shaded areas represent s.e.m. f Activity profiles for Go and NoGo cues in an example session, sorted by decision. Dashed lines represent average unsuccessful trial profiles. g Go and NoGo domain activity for all trials of the example session, with ellipses fitted to single-trial value distributions and arrows showing centroid shifts between correct and incorrect decisions. h Centroid displacement in Go and NoGo domains for all sessions, for NoGo (left) and Go (right) trials. Vectors are pointing towards activity corresponding to licking decisions (Hits and FA). Grey arrows indicate non-significant shifts (two-dimensional Kolmogorov-Smirnov test), and dotted lines represent sessions with fewer than 10 trials of one decision type. i Vectors from (h) plotted in a polar space. j Top: Measurement of angles between isoevidence lines and centroid translation vectors. Bottom: Distribution of these angles for Go (blue) and NoGo (red) stimuli.