Supplementary Figure 5: Comparison of latencies for different features in representational similarity matrices within each subregion.
From: Triple dissociation of attention and decision computations across prefrontal cortex
(a) This plot shows the same information as in bottom panels of main Fig. 4d-h, but sorted by subregion rather than by regressor. This shows more clearly the relative latencies of different variables – i.e. that spatial attention affects representational similarity earliest in each subregion; stimulus identity, attended value and accept/reject emerge roughly simultaneously, and left/right value emerges last. Whilst these findings can be used to compare latencies of these different responses, we caution that the magnitude of CPD values for different variables are not directly comparable with each other, because they depend upon the correlation structure of the design matrix. Instead, CPD values for the same variables should be compared with each other across subregions (as in main Fig. 4). (b) This plot shows the same information as in part (a), but after each regressor has been normalized to its peak value. Regressors explaining little variance have been removed from this plot, for clarity. The dashed line denotes the value at which 75% of maximal CPD is reached, which we denote as t75. Note that in both figures, CPD is estimated in sliding 200 ms bins, and is timelocked to cue onset (after the subject has saccaded to the cue). (c) Box-and-whisker plot of t75 for key variables/brain regions, estimated using different simulated observations of the noise (see Methods for details). The distributions of t75 for spatial attention do not overlap with those of attended value, accept/reject, or stimulus identity, demonstrating that spatial attention is encoded significantly earlier than other variables. Similarly, left/right value (in ACC and DLPFC) is encoded significantly later than other variables. Red lines denote median values; notches denote 95% CIs of the median; edge of boxes denote 25/75 percentiles of data; whiskers denote maximum/minimum values (excluding outliers, as estimated using MATLAB boxplot algorithm). n = 189 units (ACC), 135 units (DLPFC), 183 units (OFC).
