Bootstrapping and Empirical Bayes Methods Improve Rhythm Detection in Sparsely Sampled Data.

Area of Science:

• Genomics
• Chronobiology
• Bioinformatics

Background:

• Genome-wide expression time series are crucial for identifying circadian genes.
• High costs and effort limit data collection, impacting uncertainty assessment and statistical confidence in rhythm detection.
• Existing methods struggle with noisy or sparsely sampled time-series data.

Purpose of the Study:

• To develop a robust method for detecting rhythmic patterns in genome-wide expression time series.
• To improve the statistical confidence and accuracy of circadian gene identification.
• To address limitations in current methods caused by experimental uncertainty and sampling frequency.

Main Methods:

• Parametric bootstrapping and empirical Bayes methods for variance shrinkage.
• Development of BooteJTK, building upon the empirical JTK_CYCLE (eJTK) method.
• Integration of measurement uncertainty and rank-order information for rhythm detection.

Main Results:

• BooteJTK demonstrates rapid and accurate detection of cycling time series.
• The method shows more consistent rhythm detection than existing approaches at typical sampling frequencies.
• Application to multi-tissue datasets reveals biologically relevant tissue relationships missed by eJTK.

Conclusions:

• BooteJTK enhances rhythm detection in genome-wide expression time series by incorporating measurement uncertainty.
• The method offers improved accuracy and biological insight compared to previous techniques.
• BooteJTK is a freely available Python tool for circadian gene discovery.