Fig. 6: Unmixed components account for both overt behavior- and attention-related information.

Box plot corresponding to the distribution of the cross-validated classification accuracies of linear classifiers given by the first three demixed principal components associated with upcoming behavioral outcome (A), and with attention to target distance (TA)(B). For each parameter, the horizontal dotted line corresponds to the theoretical chance level (50% for upcoming target behavior, and 33.3% for attention to target distance). Gray boxplot shows the distribution of classification accuracies expected by chance as estimated by 100 iterations of shuffling procedure (maximal accuracy value obtained across all iterations is considered). For both parameters, only the first dPC showed accuracy values above chance level (*p < 0.05, permutation test). C Box plot corresponding to the distribution of the cross-validated classification accuracies of linear classifiers given by the first demixed principal component associated with upcoming behavioral outcome and attention to target distance using two dPCA approaches: either demixing each parameter (upcoming behavioral outcome or attention to target distance) independently or demixing each parameter from the other (two-sided Wilcoxon signed-rank test, ***p = 0.0008; ** p = 0.01) (D). Scatter plot between the normalized dot product between the first demixed components associated with attention and upcoming behavioral outcome and the behavioral gain associated with an optimal attention orientation towards the upcoming target location (Spearman correlation, ρ = −0.55, p = 0.02). All panels of this figure contained independent measures corresponding of each session (N = 18 sessions), and lower and upper box boundaries in boxplots reflect the 90th and 10th percentiles, respectively, line inside the box reflects the median and lower and upper error lines show the min and max value of each distribution, respectively.