Extended Data Fig. 1: Additional data related to huddle analysis pipeline and titration of ambient temperature.
From: Cortical regulation of collective social dynamics during environmental challenge
a. Pipeline for automated detection of huddle size. Raw frames are binarized into black and white pixels. Erosion and dilation are performed to removed tails and fecal artifacts. Edge detection is performed to identify connected groups of animals. b. Percent accuracy of detected huddle state compared to manual human annotation c. Automated Identification of huddle membership is achieved by tracking raw behavior videos with a trained neural network (Social Leap Estimates Animal Poses) to identify individual nodes and identities. Tracked poses and identities are overlayed on top of detected huddles to identify the membership. d. Example raster plot for one group demonstrating membership configurations for huddles of three throughout one behavior session. e. Frequency of group states observed when titrating ambient temperature at 20 °C, 15 °C, 10 °C, or 5 °C during thermal challenge assay (n = 6 groups of 4 individuals). f. Mean group state duration in seconds observed at 20 °C, 15 °C, 10 °C, or 5 °C during thermal challenge assay (n = 6 groups of 4 individuals). g. Rolling average (mean ± SEM) of percent time of all five group states plotted over time at 10 °C (n = 6 groups of 4 individuals). h. Rolling average (mean ± SEM) of percent time of all five group states plotted over time at 15 °C (n = 6 groups of 4 individuals). i. Rolling average (mean ± SEM) of percent time of all five group states plotted over time at 20 °C (n = 6 groups of 4 individuals). j. Individual animal’s total percent time spent in solitude in groups of four vs groups of two (n = 24 individuals from 6 groups). k. Comparison of observed huddling data versus a probabilistic model that calculates the likelihood of huddle sizes in a group of 4 based on the observed huddling in pairs. Box plots: center line–median; box limits–upper/lower quartiles; whiskers–min/max. Statistical tests: one-way (j) and two-way (e,f,k) ANOVA with Bonferroni post-hoc. *P < .05, **P < .01, ***P < .001, ****P < .0001. See Supplementary Table 1 for statistical details.
