Figure 4.

Q-learning model reveals increased sensitivity to negative feedback in experimental animals during probabilistic reversal learning. (A,B) Q-learning model estimates the update of action-value (Q-values) based on previous rewards and reward omissions. Blue and green lines represent actual choices and rewarded levers (high probability levers in case of the probabilistic condition), respectively. The purple line represents the model’s prediction of the animal’s choice. (C) No significant differences were found between groups for α_win (the degree to which previous rewards guided choices) or α_loss (the degree to which negative feedback influenced choices) under CNO administration during the non-probabilistic condition (one-way ANOVA, α_win: F1,10 = 0.291, p = 0.579 and α_loss: F1,10 = 1.257, p = 0.288). (D) While α_win did not significantly differ between groups under CNO in the probabilistic condition (one-way ANOVA, F1,10 = 0.329, p = 0.579), there was a significant difference in α_loss (one-way ANOVA, F1,10 = 8.284, p = 0.016), indicating a greater influence of reward omission in experimental animals under CNO compared to controls. No significant effects or interactions were observed with saline administration (data not shown).