Article Figures & Data
Figures
- Figure 1.
Exploit states and female sex reduce action variability during decision making. a, schematic depicting the timeline of a single trial. White squares indicate left/right spatial choice. b, An example of touchscreen responses from one animal and one session, where light purple indicates explore touches and dark purple indicates exploit touches. c, Schematic depicting the HMM and labeling explore trials along an example two-arm spatial restless bandit probability walk. Orange traces indicate the probability and choices of left side touches. Blue traces indicate the probability and choices of right side touches. Gray-shaded regions indicate HMM-labeled explore trials. d, Schematic of Euclidean distance where the distance is calculated between Touch 1 and Touch 2, Touch 2 and Touch 3, Touch 3 and Touch 4, and so on. Shown here are possible left/right touches in blue and the distance relationship from one to another represented by black lines. e, Average Euclidean distance split by state (left) and sex (right). Exploit touches and females had significantly reduced Euclidean distance. Light purple indicates distance between explore touches, and dark purple indicates distance between exploit touches. Red indicates female, and blue indicates male mice. In violin graphs, individual data points are data from one mouse averaged across all sessions. f, Schematic of Mahalanobis distance where the individual data points are measured from the overall centroid of the dataset. Shown here are possible left/right Mahalanobis clusters (light blue circles) and centroids (stars) and the Mahalanobis distance relationship from each touch (darker blue circles) in a cluster to the centroid represented by black lines. g, Average Mahalanobis distance split by state (left) and sex (right). Exploit touches had significantly reduced Mahalanobis distance. Light purple indicates Mahalanobis distance between explore touches, and dark purple indicates Mahalanobis distance between exploit touches. Red indicates female, and blue indicates male mice. h, Schematic of distance from the center of the screen where touch distance from both left and right choice apertures is measured from the midpoint of the operant screen. Shown here are possible left/right touches in blue and the distance of each from the center of the touchscreen indicated by black lines. i, Average distance from the center of the screen split by state (left) and sex (right). Explore touches were significantly closer to the center of the screen. Light purple indicates distance from the center of the screen for explore touches, and dark purple indicates distance from the center of the screen for exploit touches. Red indicates female, and blue indicates male mice. j, Schematic of response time calculation which is based on the difference between screen display and choice time (nosepoke). k, Average choice response time split by state (left) and sex (right). Exploit touches and female sex significantly reduced latency to respond. Light purple indicates response time for explore touches, and dark purple indicates response time for exploit touches. Red indicates female and blue indicates male mice. For simplicity of visualization, all plots are averages across trials and sessions, so that each individual data point plotted represents the overall average for a mouse. Significance throughout this paper is represented in the following way: *p < 0.05 and *p > 0.01; **p < 0.01 and **p ≥ 0.001; ***p < 0.001. Violin graphs depict median and quartiles of the dataset.
- Figure 2.
Previous reward reduces action variability independently from explore/exploit balance or female sex. a, Average Euclidean distance comparing rewarded versus nonrewarded trials. Touches following rewarded trials had significantly reduced Euclidean distance. Light green indicates distance between nonrewarded touches, and dark green indicates distance between rewarded touches. In violin graphs, individual data points are data from one mouse averaged across all sessions. b, Average Euclidean distance for rewarded (left) and nonrewarded (right) trials split by state and sex. Exploit touches and females had significantly reduced Euclidean distance. Red indicates female, and blue indicates male mice. c, Average Mahalanobis distance comparing rewarded versus nonrewarded trials. Touches following rewarded trials had significantly reduced Mahalanobis distance. Light green indicates Mahalanobis distance between nonrewarded touches, and dark green indicates Mahalanobis distance between rewarded touches. d, Average Mahalanobis distance for rewarded (left) and nonrewarded (right) trials split by state and sex. Exploit touches had significantly reduced Mahalanobis distance. Red indicates female, and blue indicates male mice. *p < 0.05 and *p > 0.01; **p < 0.01 and **p ≥ 0.001; ***p < 0.001. Violin graphs depict median and quartiles of the dataset.
- Figure 3.
Bouts of exploit choices are closer to each other and occupy a smaller area than bouts of explore choices. a, An example portion of a probability walk with explore trials shaded gray, illustrating state bouts and transitions between bouts. A “bout” is a period of touches within one HMM-defined behavioral state. Transitions between bouts are referenced with arrows as the state switches from explore to exploit or exploit to explore throughout. b, The average number of bouts per session split by state and sex. No significant sex difference in the number of bouts per session. Red indicates female, and blue indicates male mice. In violin graphs, individual data points are data from one mouse averaged across all sessions. c, An example 2D contour plot from Plotly Graphing Libraries fit over our nosepoke touch locations to visualize the density and range of choice responding. Small gray circles are nosepoke touches within the bout of response data. The color map corresponds with density of data points within each bin, where the darkest purple (outer bin) is the least dense contour bin, which is used to calculate the area and perimeter of the bout. d, The average area of bouts split by state and sex. Exploit touches and females had significantly reduced area. Red indicates female, and blue indicates male mice. e, Average perimeter of bouts split by state and sex. Exploit touches and females had significantly reduced perimeter. f, Schematic depicting centroid shifts, where the Euclidean distance between two successive Mahalanobis centroids is calculated. Stars indicate example centroids associated with bouts, and black lines indicate the distance calculations between those centroids. g, Centroid shifts split by state and sex. Centroid shifts were significantly smaller for exploit bouts. Red indicates female, and blue indicates male mice. *p < 0.05 and *p > 0.01; **p < 0.01 and **p ≥ 0.001; ***p < 0.001. Violin graphs depict median and quartiles of the dataset.
Tables
Variable Coded as GLMM series Dependent variable Figure Beta Coeff. P value (state|mouseID) Random Euclidean dist_mm 1e 10.42632 3.52 × 10−11 Sex Categorical Euclidean dist_mm 1e 2.668432 0.012418 State Categorical Euclidean dist_mm 1e 4.479342 0.000118 Sex × state Interaction Euclidean dist_mm 1e −1.63685 0.112654 (state|mouseID) Random Mahalanobis mahalanobis 1g 42.47032 1.27 × 10−29 State Categorical Mahalanobis mahalanobis 1g 6.132201 2.03 × 10−6 (state|mouseID) Random Distance From Center cent_dist_mm 1i 24.75777 1.70 × 10−21 Sex Categorical Distance From Center cent_dist_mm 1i 0.299544 0.766587 State Categorical Distance From Center cent_dist_mm 1i −4.72546 5.15 × 10−5 Sex × state Interaction Distance From Center cent_dist_mm 1i 0.767197 0.448969 (state|mouseID) Random Euclidean dist_mm 2a,b 11.7435 1.04 × 10−12 Sex Categorical Euclidean dist_mm 2b 2.779354 0.009519 State Categorical Euclidean dist_mm 2a −0.30298 0.764276 Reward Categorical Euclidean dist_mm 2a,b −6.96119 9.88 × 10−8 Sex × state Interaction Euclidean dist_mm 2a,b −2.17184 0.038554 (reward|mouseID) Random Mahalanobis mahalanobis 2c,d 29.75193 3.78 × 10−38 State Categorical Mahalanobis mahalanobis 2c 4.383663 0.000181 Reward Categorical Mahalanobis mahalanobis 2c,d −4.01022 0.000364 (1|mouseID) Random Local centroid distance loc_dist2_lastLoc 3g 10.73884 1.18 × 10−11 Sex Categorical Local centroid distance loc_dist2_lastLoc 3g −0.01149 0.99088 State Categorical Local centroid distance loc_dist2_lastLoc 3g 8.964287 4.03 × 10−19 State × sex Interaction Local centroid distance loc_dist2_lastLoc 3g 2.078314 0.03772 (1|mouseID) Random Average Bout Area contour_area 3d 5.058701 1.95 × 10−5 Sex Categorical Average Bout Area contour_area 3d 4.731083 4.57 × 10−5 State Categorical Average Bout Area contour_area 3d −2.661 0.00781 State × sex Interaction Average Bout Area contour_area 3d −2.87523 0.004051 (1|mouseID) Random Average Bout Perimeter contour_perimeter 3e 13.83109 1.05 × 10−14 Sex Categorical Average Bout Perimeter contour_perimeter 3e 3.643466 0.000928 State Categorical Average Bout Perimeter contour_perimeter 3e −7.16055 8.93 × 10−13 State × sex Interaction Average Bout Perimeter contour_perimeter 3e −1.53537 0.124742 (state|mouseID) Random Response Latency RT 1k 8.731039 1.23 × 10−9 Sex Categorical Response Latency RT 1k 3.891523 0.000528 State Categorical Response Latency RT 1k 4.54362 8.54 × 10−5 (1|mouseID) Random Learning Rate alpha N/A 11.52081 1.76 × 10−14 Sex Categorical Learning Rate alpha N/A 2.512188 0.017736 Alpha Continuous Learning Rate alpha N/A −2.01035 0.045508 (1|mouseID) Random Bout number num_explore 3b 12.12988 6.87 × 10−13 Sex Categorical Bout number num_explore 3b −2.1784 0.037647 (1|mouseID) Random Bout number num_exploit 3b 10.75028 1.23 × 10−11 Sex Categorical Bout number num_exploit 3b −2.36733 0.024802 Table including equations, variables, figure references, beta coefficients, and p values for all included statistics.
In this issue
