Fig. 4: Temporal relation between neural activity and behavior.

a, Schematic of video to activity prediction with different temporal offsets. Comparison across shifted time windows. b, Comparison of prediction accuracy across different temporal shifts averaged by brain area. For each area, the explained variance across different temporal shifts is shown as a line, normalized to the explained variance at zero temporal shift. Lines correspond to mean across sessions, colored by brain region; shaded area corresponds to the s.e.m. across sessions. c, Brain-wide spatial map of fraction of neurons with positive (left) and negative (right) best temporal offsets. Voxels are of size 300 µm squared in the sagittal plane and span the brain in the third direction. Color represents the proportion of neurons with positive time-offset (left) and negative time-offset (right) within each voxel. For visualization a 3 × 3-voxel median filter was applied on the heatmaps. Scale bar: 0.5 mm. d, Heatmap of best time-offsets for ALM neurons. Each voxel is 150 µm squared in the coronal plane and spans the brain in the third direction. Color corresponds to the average of best temporal offsets within that voxel. For visualization a 3 × 3-voxel median filter was applied on the heatmap. Scale bar: 1 mm. e, Comparison of explained variance across different temporal offsets as a function of cortical depth, taken as distance along the dorsal–ventral axis shown in d. Each row is normalized by its minimal and maximal values. Spearman rank correlation between cortical depth and best time-offset is R = 0.1, P = 0.0014, n = 1,059 neurons.