Extended Data Fig. 10: Bilinear encoding models allow value to multiplicatively modulate all event-related temporal kernels.

a. Schematic of the encoding model used to estimate how trial-by-trial fluctuations in relative side value (Q_I−Q_L, where Q_I and Q_C are the action values for the lever ipsilateral and contralateral to the imaging field, respectively) or relative chosen value (Q_Ch−Q_U, where Q_Ch and Q_U are the action values for the chosen and unchosen lever, respectively) influences responses to each task event. For each neuron, GCaMP6f fluorescence was estimated as the sum of all event kernels multiplied by the gain coefficients (an offset gain as well as gains for relative chosen value (Fig. 6a–c) or relative side value (Fig. 6d–f)). b. Fluorescence (gray) and estimated fluorescence (green) from the same neuron during the outcome period on 3 trials with different upcoming relative chosen values (labeled on bottom). c. Schematic of the 2-step iterative fitting procedure (see Methods for more details). d. The proportion of neurons significantly encoding current relative chosen value during all task events for males and females. Significantly more neurons encoded current trial relative chosen values in females compared to males during the nose poke event (χ²-test, χ²(1,N = 756) = 4.04, p = 0.04) and significantly more neurons encoded current trial relative chosen value in males compared to females during the outcome event (χ²-test, χ²(1,N = 756) = 5.30, p = 0.02). There were no sex differences in the encoding of relative chosen value for any other task events (χ²-test, p > 0.05). e. The proportion of neurons significantly encoding current relative side value during the nose poke, lever presentation, and outcome periods and the upcoming relative side value during rewarded and unrewarded outcomes. Significantly more neurons encoded upcoming relative side value in males compared to females during rewarded outcomes (χ²-test, χ²(1,N = 756) = 5.35, p = 0.02).

Source data