CycleVI: Isolating cell cycle variation with an interpretable deep generative model

Abstract

MotivationCell cycle progression is a dominant source of variation in single-cell RNA sequencing (scRNA-seq) data, often obscuring other transcriptional signals of interest. Several methods have been developed to infer continuous cell-cycle phase from transcriptomic data, but their estimates tend to be unstable when proliferation is intertwined with other biological processes or technical sources of heterogeneity. ResultsWe present CycleVI, a deep generative model that disentangles cell cycle-driven variation from other signals in scRNA-seq data using a partitioned latent representation with a dedicated circular subspace. CycleVI accurately infers a continuous cell cycle phase, validated against orthogonal protein-level measurements, and yields a residual latent space free of cell cycle artifacts. This disentangled representation helps resolve biological processes intertwined with the cell cycle, clarifying hematopoietic differentiation and preserving drug-response signals better than standard cell cycle regression. By iso-lating cell cycle-related variation rather than removing it, CycleVI provides a principled framework for analyzing cellular heterogeneity in proliferating systems. Availability and ImplementationCycleVI is available at www.github.com/jeuken/CycleVI.