Statistical Methods for Detecting Circadian Rhythmicity and Differential Circadian Patterns with Repeated Measurement in Transcriptomic Applications

Circadian analysis via transcriptomic data has been successful in revealing the clock output changes underlying many diseases and physiological processes. Repeated measurement design in a circadian study is prevalent, in which the same subject is repeatedly measured over time. Several methods are currently available to perform circadian analysis, however, none of them take advantage of the repeated measurement design. And ignoring the within-subject correlation from the repeated measurement could result in lower statistical power. To address this issue, we developed linear mixed model based methods to detect (i) circadian rhythmicity (i.e., Rpt_rhythmicity) and (ii) differential circadian patterns comparing two experimental conditions (i.e., Rpt_diff). Our model includes a subject-specific random effect, which will account for the within-subject correlation. Via simulations, we showed our method not only could control the type I error rate around the nominal level, but also achieve higher statistical power compared to other methods that cannot model repeated measurement. The superior performance of Rpt_rhythmicity and Rpt_diff were also demonstrated in two real data applications, including a human restricted feeding data and a human sleep restriction data. An R package for our methods is publicly available on GitHub to promote the application of our methods.