Extended Data Fig. 12: Two-action learning analyses supplementing Fig. 4.
From: Dynamic behaviour restructuring mediates dopamine-dependent credit assignment
a-c, Emergence of distinct type of learning curves specifically amongst action sequence learners. a, Learning curves of single action learners. Normalized frequency indicates that frequencies were normalized to the highest frequency achieved in a session for each ChR2-YFP animal. b, Open field inter-action intervals of T1/T2 pairs chosen in action sequence learning. Same colour codes in (c, g-m). c, Overlaying learning curves of single action learners (grey, dashed lines) with action sequence learners (colour coded as in (b)). d-f, Linear relationships assessed in single action learners. d, No significant linear relationship between baseline action frequency and reinforced frequencies in Session 1 (top) and 3 (bottom). e-f, The number of action (clusters) transitioning into (e) and out of (f) target at baseline is negatively related to the fold change in frequency from baseline to last session. F-test (F(1, 13) = 83.31(e), 17.88(f); p = 5*10−7 (e), 1.0*10−3 (f). g-l, Action-to-Target (g, i) and Target-to-Action (h,j) and differential T2-T1 (k,l) transitioning parameters are not related to sessions to criterion frequency parameter. F-test: F(1, 12) = 1.11(g), 0.0053(h), 0.037(i), 0.014(j), 1.49(k), 0.048(l); p = 0.31(g), 0.94(h), 0.51(i), 0.69(j), 0.24(k), 0.83(l). Some data points overlap. m, No significant linear relationship between implant location and sessions to criterion. F-test: F(1, 12) = 0.11, p = 0.75). Some data points overlap (g-m). n, Lack of sequence reinforcement in single action reinforcement. Learning curves differ across time and group (2-way ANOVA – F(2, 54) = 143.3, p < 0.0001). Plot indicates mean +/− SEM. n = 15 and 14 ChR2-YFP animals (biological replicates) in single action reinforcement and action sequence reinforcement (past criterion frequency) experiments, respectively. n.s. – not significant.
