Dissertation Defense: Thien Pham
"Statistical Methods for the Design and Analysis of Circadian Omics Studies", School of Public Health Department of Biostatistics and Health Data Science.
Advisor and Committee Chair: George Tseng
Abstract:
Circadian rhythms coordinate daily physiology and influence health and disease. Advances in high-throughput
technologies have expanded circadian omics studies across tissues, species, and molecular modalities. Limited
sample sizes and uneven sampling times require careful power analysis and study design. Comparisons
between conditions help distinguish conserved from altered rhythmic programs and assess how well animal
models recapitulate human circadian biology. Integrating evidence across tissues further reveals the temporal
organization of these programs across organs and their conservation across species. This dissertation develops
three statistical frameworks to address these connected needs.
In Chapter 2, we develop Simulation-based Circadian Power (SCP), a framework for power analysis and sample
size planning in circadian omics studies. SCP uses pilot-derived distributions of amplitude, phase, sampling time,
and noise to capture transcriptome-wide signal heterogeneity. It estimates power under false discovery rate
control for rhythmicity detection and two-group differential circadian analyses, supporting both cosinor and non-
cosinor methods. Bootstrap analysis quantifies uncertainty arising from pilot data. SCP also provides 160
curated circadian transcriptomic datasets spanning more than 100 tissues across multiple species to support
pilot calibration and study design.
Determining whether rhythmic programs are conserved requires more than testing for differences between
conditions. In Chapter 3, we introduce BayesCR (Bayesian Comparative Rhythmicity), a Bayesian framework for
evaluating cross-condition congruence at the gene, pathway, and genome-wide levels. Posterior rhythmicity
probabilities and Bayesian differential and equivalence tests identify rhythm-conserved (RC), rhythm-gained
(RG), and rhythm-lost (RL) genes and classify phase relationships among rhythm-conserved genes as
conserved (PC), advanced (PA), or delayed (PD), with Bayesian false discovery rate control. Congruence scores
quantify conservation and alteration across pathways and the transcriptome. Applications to human, baboon,
and mouse transcriptomes identify conserved and altered programs, including systematic phase shifts in
human–baboon and human–mouse comparisons. BayesCR provides the first statistical framework to formally
quantify cross-species circadian congruence, helping assess which circadian mechanisms in animal models may
translate to humans.
Understanding the pan-tissue temporal organization of rhythmic gene activity requires integrating evidence
beyond pairwise tissue comparisons. In Chapter 4, we developed an integrative Bayesian model that combines
information across multiple tissues to estimate each gene’s canonical phase, a common reference peak time
inferred jointly across tissues. The model characterizes pan-tissue rhythmicity by assessing both the breadth of
rhythmic expression across tissues and the coordination of tissue-specific rhythms around this canonical phase.
Applications to baboon and human transcriptomes examine the temporal organization of circadian programs
across organs and their conservation across species.
Public health significance: This dissertation advances circadian study design and strengthens translational research by assessing how well animal models reflect human circadian biology. Together, these methods provide a statistical toolkit for investigating circadian regulation in health and disease and informing future research on disease prevention and chronotherapy.