Extended Data Fig. 7: Linking statistical hierarchy of pattern elements to dynamics.

a, Three example intermediate-level clusters of motor elements (threshold of 0.4 as in Fig. 3i, different animal), overlaid on the average aligned colour image for the dataset (216,160 images). The clusters are mostly composed of chromatophores of a single colour: cluster 1 (red) is light; clusters 2 and 3 (green and blue) are dark. b, The dynamics of a 60-min dataset, projected onto the first three principal components (48% variance explained, n = 1,437 chromatophores, 52,040 samples). A cluster activity direction can be defined in principal component space by projecting the cluster identity vector (vector of length = number of chromatophores, with 1s assigned to chromatophores in a cluster, otherwise it is assigned 0), onto the principal components. The coloured lines show the cluster activity directions for the three clusters in a. Projecting the dataset onto these directions shows the expansion strength of the cluster at different times. The images corresponding to the times of lowest and highest strengths are shown to the left and right, respectively. c, Full distribution of expansion strengths, projecting all time points onto cluster activity directions. In this dataset, cluster 2 is often expanded, whereas clusters 1 and 3 are rarely expanded. a.u., arbitrary units.