Extended Data Fig. 6: Comparison of AUC- and deconvolution-based measures of neuronal encoding for different cell types.
a–c, Each panel plots encoding scores derived from the AUC method (y axis) against those derived from deconvolved calcium traces (x-axis), for reward cells (a), reward-approach cells (b) and screen cells (c). Across all comparisons, we observe a strong correlation between the two methods, indicating consistency in the estimation of encoding score (n = 1025 reward cells, 558 reward-approach cells, and 674 screen cells). The inset in each panel shows R2 across all mice (n = 7 mice). Bar graph and error bar in inset of a–c show mean ± s.e.m. d–g, Temporal dynamics of percentage of reward-encoding cells using AUC-based analysis. d,e, Using an AUC-based cell identification method, we quantified the percentage of neurons encoding reward and examined their relationship with session number (day) (d) and behavioural performance (e). f, Correlation analyses across all mice (n = 7 mice) show that the percentage of reward cells is negatively correlated with session number and only weakly correlated with performance. g, Results from our linear model further support this observation, demonstrating that the evolution of reward-cell recruitment is better explained by session number rather than performance—consistent with findings presented in the main manuscript using averaged activity obtained using deconvolved traces (n = 7 mice). Bar graphs and error bars in f,g show mean ± s.e.m.
