Fig. 5: Multiplexing of task information with varying stability within single RSC neurons.

a Average performance of support vector machine classifiers on the activity of single cells that significantly encode two task variables: context and motor choice (top row), context and post-decision outcome (middle row), and motor choice and post-decision outcome (bottom row) in day 1. Additional cells with multiplexing decoding of different pairs of variables are shown in Supplementary Figs. 11 and 12. Note the better decoding performance of models trained in day 1 and tested in days 2 and 4 for context and outcome variables. b Example cells with significant encoding of context and motor choice (cell 1, top), context and post-decision outcome (cell 2, middle), and motor choice and post-decision outcome (cell 3, bottom) in day 1. Solid line, mean; shaded area, 95% CIs. c Average decoding performance of the two task variables indicated in each plot within individual cells (mean ± s.e.m.; n = 43, 23, 40, 35, 40 cells in days 1–5 [context-motor]; n = 48, 31, 42, 44, 44 cells in days 1–5 [context-post-decision outcome]; n = 55, 33, 51, 51, 53 cells in days 1–5 [context-post-trial outcome]; n = 52, 30, 49, 47, 49 cells in days 1–5 [motor-post-decision outcome]; n = 55, 31, 51, 49, 51 cells in days 1–5 [motor-post-trial outcome]). Asterisks denote significantly different decoding performance, as measured by different fit (*p < 0.05, **p < 0.01, ***p < 0.001; two-sided F-test calculated by LME models; see “Methods” section for details). Note the higher decoding performance of the classifiers decoding context and outcome when compared to motor choice. For all panels: s trial start, d decision point, e trial end. Source data for C are provided as a Source Data file.