Figure 3—figure supplement 1.

Figure 3—figure supplement 1. One time-point artifact or one noisy neuron fluorescence trace is sufficient to corrupt the GC algorithm.(A) Mean image of a 30-somite stage larval zebrafish Tg(s1020t:Gal4; UAS:GCaMP3), dataset f3t1 in De Vico Fallani et al., 2014, with neurons selected for analysis shown in color. (B) Fluorescences traces of neurons of (A). Top: A motion artifact consisting of a drop in fluorescence at a single time point is present (star). Bottom: the motion artifact is corrected as being the mean of the two surrounding time points. (C) Directed graph resulting from the bivariate (left) and multivariate (right) GC analysis, before (top) and after (bottom) motion artifact correction. Before the correction, the neuron with the smallest activity appears to drive many other neurons, perhaps due to its lower SNR. This spurious dominant drive disappears once the motion artifact is corrected, demonstrating that GC is not resilient to artifacts as small as one time point out of one thousand. (D) Mean image of a 30-somite stage larval zebrafish Tg(s1020t:Gal4; UAS:GCaMP3), dataset f5t2 in De Vico Fallani et al., 2014, with neurons selected for analysis shown in color. The fluorescence traces (E) of neurons displayed in red exhibit activity patterns different from the oscillatory pattern expected in motorneurons. GC analysis was run after removal of these neurons. (F) Directed graph resulting from the bivariate (left) and multivariate (right) GC analysis, before (top) and after (bottom) removing neuron 13 and neuron 21.