Dataset contains timeseries data used for testing period analysis methods. The full details of the data are described in the actual paper, here there is only a short summary of methods generation.
The synthetic test signals comprise:
- a pure cosine of known frequency, and hence known period;
- a pulsed waveform which comprises pulses of a Gaussian waveform with a standard deviation of (period/7);
- a double pulsed waveform which comprises two periodic Gaussian waveforms;
- simulated clock data generated using a delayed negative feedback loop (DNFL) model. Here we use parameter sets identical to those used in Costa MJ et al. Biostatistics 2013 to produce two sets with moderate level of asymmetry or a moderate shoulder.
Once the basic time series had been generated, noise was added. The noise was additive and was either uniform noise or walking noise. Uniform noise is drawn from a uniform distribution and the amplitude of the noise is defined as a percentage of the amplitude of the original time-series. This would be characteristic of the noise in a measurement system. Walking noise is additive and uniform, but this time the distribution of the noise is restricted so that the current data point lies within a limited range of the previous data point.
To assess applicability of the methods to the analysis of real biological systems, we also tested data sets obtained by in vivo imaging of transgenic Arabidopsis thaliana plants.
Zielinski, Tomasz; Millar, Andrew J. (2019). Timeseries used in Zielinski et al. Strengths and Limitations of Period Estimation Methods, [dataset]. University of Edinburgh, School of Biological Sciences and SynthSys. https://doi.org/10.7488/ds/2636.