Extended Data Fig. 8: Effect of past interactions, relative rank and reward on neuronal response.
From: Frontal neurons driving competitive behaviour and ecology of social groups
a, Venn diagram depicting the number of neurons encoding the animals’ ordinal (i.e., first vs. second vs. third vs. fourth to reach reward entry zone) competitive success overlapping with neurons encoding binary (i.e., first two vs last two to reach reward entry zone) competitive success (p<0.01, FDR corrected for multiple epochs). Most neurons that encoded the animals’ ordinal and binary competitive success overlapped (n=120; χ2(1)=30.7, p=3x10-8; Chi-Square test). b, Decoding accuracy for ordinal competitive success for all task-modulated neurons (n=560; p<0.05; Permutation tests). Decoding accuracy gradually increased to a peak of 63.2±1.5% prior to reward zone entry (p<0.001, Permutation test). c, Venn diagram depicting the degree of overlap between neurons encoding the animals’ ordinal relative rank (i.e., first vs. second vs. third vs. fourth) and binary (i.e., two highest vs two lowest) relative rank (p<0.01, FDR corrected for multiple epochs). Neurons encoding the animals’ ordinal relative rank almost entirely overlapped with those that encoded the animal’s binary relative rank (n=122 overlap; χ2(1)=29.3, p=6.24x10-8; Chi-Square test). d, Using all task-modulated neurons (n=560), decoding accuracy for ordinal relative rank were significantly higher than chance over the course of the trials (p<0.0001; Permutation tests). e, Decoding accuracies for relative rank and reward size using only neurons that encoded relative rank (Left, n=87) or reward size (Right, n=156), respectively. f, Recent history effects. Top, example trial sequence and count of cumulative matching trials in succession. Bottom, number of ‘matching’ trials in succession when comparing wins vs losses at trial t (Z=1.013, p=0.31; Rank-sum) suggesting transient, short-lasting successions of wins and losses. N=4966 total trials. g, Venn diagrams depicting the number of neurons encoding competitive success in the past trial t-1 and their overlap with neurons encoding competitive success (Top) or relative rank (Bottom) in the present trial t (p<0.01, two-way ANOVA, FDR corrected for multiple epochs). Most neurons that encoded the animals past success (5.1%, n=54) overlapped with those that encoded their current success (n = 34; χ2(1)=1.31, p=0.25; Chi-Square test) but were largely distinct from those that encoded the animals’ relative rank (n=9; χ2(1)=12.2, p=0.00046; Chi-Square test). h, Using all task-modulated neurons (n=560), decoding accuracies for the previous trial competitive outcome (t-1) were not significantly different than chance (p>0.05; Permutation tests). Peak decoding accuracy for past success was 57.2±2.3% (H0=50% chance performance; p>0.05; Permutation tests). i, Using all task-modulated neurons (n=560), decoding accuracies for the animal’s current success (t) contingent on the previous trial’s competitive outcome (t-1) was significantly higher than chance (p<0.01; Permutation tests). These neurons predict the animal’s current success contingent on their past outcome with accuracy of up to 41.3±3.1% prior to trial onset (H0=25%, p<0.001; Permutation test). j, Using all task-modulated neurons (n=560), considering succession of wins or losses (behavioural states), decoding accuracy for the animals’ behavioural state prior to gate opening was 66.2±2.0% (H0=50% chance performance; p<0.01, Permutation test). k, Using all task-modulated neurons (n=560), peak decoding accuracy for the animals upcoming success contingent on their prior behavioural state prior to gate opening was 64.7±3.7% (H0=25% chance performance; p<0.001; Permutation test). l, GLMs were used to quantify the contribution of past behaviour to neural population response for competitive success (Methods). Time points in which the fraction of explained variance for the interaction between terms was higher than expected from chance (p<0.01; Permutation test). Trials are aligned to the gate opening or time point at which the recorded animal reached the reward zone. Shaded areas denote mean±95%CI. N=500 bootstrapped samples for all decoding results.
