Coherence analysis of noisy simulated data and comparison with spike inference methods. White noise was added to the simulated fluorescence generated using parameters for . The amplitude of the noise was varied from 0 to 5 times the height of a fluorescence peak because of a single spike, plus the special case where there is only noise and no signal (∞ symbol). At each noise amplitude, we used 3 methods, coherence analysis and spike inference based on either thresholding or template-matching, followed by circular statistics to estimate rhythmicity and phase. We repeated the analysis 100 times, each a trial average of 3 traces, to generate a mean and SD of the estimates. *A*: estimate of rhythmicity, measured as coherence magnitude or *r* value from circular statistics, as a function of noise amplitude. Because both coherence magnitude and *r* value are bounded by 0 and 1, they are plotted with the same *y*-axis. The |*C*|_{P = 0.05} value is indicated by a gray line. *B*: for coherence, fluorescence was classified as rhythmic at *P* value of 0.05 if coherence magnitude > |*C*|_{P = 0.05}. For *r* value, fluorescence was classified as rhythmic at *P* value of 0.05 using the Rayleigh test. *C*: estimate of phase as a function of noise amplitude. All error bars show ±SD to display the spread of the rhythmicity and phase estimate.

## PubMed Commons