Effects of stimulation ranged from very large and reliable transients (Figure \ref{489066}Ai) to undetectable changes (Figure \ref{489066}Aiii). In between those extremes, we observed transients of variable size and reliability (Figure \ref{489066}B). To our surprise, we could also detect clear inhibitory responses (Figure \ref{489066}Aii). It was possible because —at least in some cell types— fluctuations in baseline activity occasionally elevated GCaMP levels during the experiment (see Discussion and Supplementary Figure \ref{998826}). Therefore, even though hyperpolarization below resting potential is likely not detectable through calcium imaging, we could detect inhibition from an excited state as a dip in the fluorescence trace.
Since no single characteristic of the responses could adequately capture their variety, variability and complexity, we chose to characterize the transients by using a battery of statistics reflecting response amplitude, shape, reliability and stimulus sensitivity (see Figure \ref{489066}C, Supplementary Figure \ref{695432}, and Materials and Methods). Responses of control pairs with non-overlapping processes were then used to form the null-hypothesis distributions of two metrics that capture response amplitude and reliability (see Figure \ref{489066}D). For every data point, the Mahalanobis distance (see Methods) to the null distribution was computed and used as a connection strength metric in summary diagrams like Figure \ref{495744} and \ref{704309}. Non-overlapping pairs usually showed no fluctuations upon stimulation, and when they did, they were small and unreliable (see Supplementary Figure \ref{822823}), likely reflecting indirect effects. Not surprisingly, responses were always detected with same-cell-type-stimulation controls (see Figure \ref{489066}D).