Inferring circadian phases and quantifying biological desynchrony across single-cell transcriptomes

ABSTRACT: Single-cell RNA sequencing (scRNA-seq) offers a unique opportunity to dissect circadian heterogeneity, yet accurately inferring circadian phase remains challenging due to natural biological fluctuations, technical noise, and low sequencing depth. Here, we introduce scRitmo, an unsupervised, model-based approach that jointly infers single-cell circadian phases and their posterior uncertainties. Using a simulation framework, we demonstrate that the non-uniform expression of core clock genes creates phase "attractor zones" that bias inference at low sequencing depths, and show that incorporating a broader set of rhythmically expressed genes allows scRitmo to overcome this limitation. We applied scRitmo to diverse datasets to unravel cell-type-specific circadian organization. Benchmarking across multiple murine scRNA-seq datasets demonstrates that scRitmo outperforms existing methods. We further validated our approach using a novel dataset composed of deeply sequenced unsynchronized fibroblasts, showing that transcriptomic phases accurately reconstruct temporal ordering of the corresponding protein levels. Beyond point estimates, scRitmo provides a measure of phase confidence through its posterior circular standard deviation. We show that this quantity is a robust predictor of inference quality: in both simulations and diverse murine datasets, cells with lower posterior uncertainty consistently exhibit significantly lower absolute errors relative to external time, allowing for principled quality control of single-cell predictions. Crucially, scRitmo enables the quantification of biological (de)synchrony by disentangling true phase dispersion from technical noise. We validated this capability using both synthetic datasets with known ground-truth desynchrony and time-series single-molecule RNA FISH (SABER-FISH) data. In the latter, scRitmo accurately recovered the progressive accumulation of phase heterogeneity following synchronization, confirming its ability to track dynamic changes in population synchrony. Applying this capability to Drosophila brains, scRitmo captured the expected increased desynchrony in constant darkness compared to light-dark cycles, effectively quantifying the biological loss of synchrony in the absence of Zeitgebers. Together, scRitmo provides a rigorous framework for high-precision phase inference, establishing a principled approach to decouple technical noise from biological variance and enabling the quantitative study of circadian (de)synchrony as a fundamental property of multi-scale temporal organization.

ORGANISM(S): Mus musculus

PROVIDER: GSE325045 | GEO | 2026/03/30

REPOSITORIES: GEO