Different, same, or arrhythmic: an empirically informed benchmark of differential rhythmicity methods

Abstract

Circadian rhythms represent a key biological dynamic across all kingdoms of life. These cell-autonomous rhythms generated by a gene-regulatory network regulate processes in a time-of-day-specific manner primarily via transcriptional rhythms in a significant fraction of genes. The controlled experiment paradigm can be used to study causal effects at the molecular level using transcriptomic RNA-seq data even in this dynamic scenario. Multiple tools for the analysis of such data, termed differential rhythmicity (DR) analysis, are available but guidance on method choice remains limited. Here, we present a systematic benchmarking study of 12 different methods from 5 packages using a synthetic model of RNA-seq count data that incorporates both empirically estimated mean-dispersion trends and concomitant differential expression changes. By linking method performance to experimental conditions and analytical goals, we provide practical and simple rules of thumb that will help biologists choose appropriate DR methods and extract reliable biological insight from time series RNA-seq data.