Go to https://circada.shinyapps.io/circada-s/
This app allows you to generate rhythmic time series with specific parameters. Then you can apply 5 different analysis techniques to your time series data and find out how well they can detect your specific parameters, such as period and phase.
Why would you want to do this? Maybe you are planning an experiment and wonder if you would be better to collect samples more frequently and test over fewer days, or should you spread your samples out over more days? Perhaps you have collected data that looks possibly rhythmic, but it seems quite noisy or variable, and you wonder what technique might be best for your data analysis leading to an estimate of circadian period. There could be many reasons.
Tips:
Navigating: The gray left-hand panel controls the characteristics of the generated time series. After you alter settings, you will see changes in the graph on the top right, under the tab “Overview”.
When you change settings, remember to click Go.
Sampling interval: The sampling intervals available in the CIRCADA-S app were chosen to give a range of lengths that are most compatible with the discrete wavelet transform (DWT).
Length of time series: You can vary this between 3 to 15 days. Studies of less than 3 days duration would be analyzed by different techniques.
Standard deviation of periods: If more than one time series is to be generated, periods are randomly chosen for each time series from the normal distribution with the indicated mean and standard deviation (SD). If you want all samples to have the same period, set SD to zero.
Amplitude: We define amplitude to equal half of the peak-to-trough height of the waveform.
Peak location: This value is how far into the total cycle the peak will occur. For example, for a period of 24h, the peak will occur at time 12h and every 24h thereafter if peak location is 0.5; it will occur at time 0, 24, 48, and so on, if peak location is 0. Due to the periodicity, peak location 0 and 1 yield roughly the same result.
Waveform: To compare how well the methods work for different waveforms (periodic patterns), the app has a choice of sinusoid, square wave, and an asymmetric triangular wave. Take a look at each type, then return to sinusoid.
Noise characteristics: Gaussian noise is normally distributed and its frequency spectrum will tend to be fairly uniform. Pink (1/f) noise has more low frequency noise than Gaussian, and Brown (1/f^2) noise will have even more. Noise “scaled by the signal” is generated by taking Gaussian noise and multiplying pointwise by the signal.
Trend: The added trend can be set to none, linear, or quadratic. The slope of the added trend is chosen randomly and could be either positive or negative.
Number of samples: Because each time series has randomly generated noise (and also a randomly generated period if SD of periods is greater than zero), we typically want to test batches of these time series to observe the range of errors in period or phase estimates. The analysis for all methods is run automatically on all generated time series, so can take some time if a very large number of samples is chosen. We recommend restricting to 1000 or less, to avoid a long computational runtime. For purposes of exploring the app, you could set the # of samples to 100 and click Go.