A Bayesian framework for genome-wide circadian rhythmicity biomarker detection

Circadian rhythms are endogenous $\sim $24-h cycles that significantly influence physiological and behavioral processes. These rhythms are governed by a transcriptional-translational feedback loop of core circadian genes and are essential for maintaining overall health. The study of circadian rhythms has expanded into various omics datasets, necessitating accurate analytical methodology for circadian biomarker detection. Here, we introduce a novel Bayesian framework for the detection of circadian rhythms in genome-wide transcriptomic applications that is capable of incorporating prior biological knowledge and adjusting for multiple testing issue via a false discovery rate (FDR) approach. Our framework leverages a Bayesian hierarchical model and employs a reverse jump Markov chain Monte Carlo technique for model selection. Through extensive simulations, our method, BayesCircRhy, demonstrated favorable FDR control over competing methods, robustness against heavier-tailed error distributions, and better performance compared with existing approaches. The method's efficacy was further validated in two RNA-sequencing data, including a human-restricted feeding data and a mouse aging data, where it successfully identified known and novel circadian genes.