INTRODUCTION: RAIN is a nonparametric method developed to detect circadian activity in data which underlying period is not well-approximated by a sine. While widely used in research on gene expression, its formulation makes its use limited to sparsely sampled data, e.g. once every hour. At the same time, a method capable of detecting non-sinusoidal rhythmicity and assigning statistical significance to the given period would be a valuable tool in different fields connected to chronobiology, like actigraphy—which stands for measurement of movement intensity—due to asymmetrical nature of human sleep/wake cycle.

AIM(S): To develop a method for nonparametric estimation of asymmetric periodicity.

METHOD(S): We present an extension of RAIN, which changes the way it groups datapoints and the way it handles multiday recordings by aggregation of points in windows and estimating net p-value for a multiday recording, while preserving probability distribution underlying the zeroth hypothesis in the original estimator. Such reformulation is carried to make the algorithm usable on data sampled with higher sampling rates.

RESULTS: The benchmarks on synthetic data exhibit correct detection of underlying rhythmicity, while allowing use of higher sampling rates. In comparison with the cosinor method (which stands for fitting a sine using the least squares optimization), the presented version of RAIN exhibits better detection of asymmetric oscillations in data.

CONCLUSIONS: Presented rhythm detection methodology is a step forward in comparison with original RAIN, allowing nonparametric detection of asymmetric periods in biological data with higher sampling rates, however it needs further development in order to make it more computationally efficient and robust to rhythmicity changes across days during a multiday recording.