Fig. 6.

Model performance on sample dynamics from evaluation set. This is a set of well-understood dynamics reserved for model evaluation. All simulation scale bars are 500 μm. a Ablation of a rectangular region in vivo⁴ and b ablation in silico (also see Supplementary Movie 7). In both cases, the resulting patterns are characterized by stripes with normal width but lost directionality. Ablation is simulated at 40 dpf (when our pattern seems to most closely resemble the first image in the timeline (a)) by removing all melanophores and xanthophores and 20% (randomly selected) of loose and dense iridophores (motivated by the indirect effects of laser ablation on iridophores described in ref. ⁷) in a rectangular region 1000 μm long (and the full height of the domain). c Puma involves a strong reduction of M birth (scale bar is 2 mm)³⁴; d puma is simulated by reducing M birth by ≈85%. e Choker lacks the typical central strip of I^d at 21 dpf; instead, these cells arise randomly at a later time¹¹. f Choker is simulated by removing the I^d strip from our initial condition and introducing 15 I^d cells at random locations at 31 dpf (see Supplementary Movie 6). g M CV averaged across 100 wild-type simulations and 80 pfeffer simulations (error bars denote standard deviation) plotted versus time; empirical measurements of CV range from 100 to 30 during wild-type development based on our approximation of data in ref. ³³, and higher CV corresponds to reduced xanthophores⁶ (as shown, we find higher CV values in pfeffer, which lacks xanthophores). Image a reproduced from Yamaguchi et al.⁴ with permission; Copyright (2007) National Academy of Sciences, U.S.A. Image c reproduced from Parichy et al.³⁴ with permission from Elsevier; Copyright (2003) Elsevier Science, U.S.A. Image e reproduced with minor adaptation from Frohnhöfer et al.¹¹ and licensed under CC-BY 3.0 (http://creativecommons.org/licenses/by/3.0); published by The Company of Biologists Ltd