Modeling Bi-modality Improves Characterization of Cell Cycle on Gene Expression in Single Cells

• DOI: 10.1371/journal.pcbi.1003696
• Publication Date: 2014-07-17T18:42:35Z
• Authors: Andrew McDavid; Lucas Dennis; Patrick Danaher; Greg Finak; Michael Krouse; Alice Wang; Philippa Webster; Joseph Beechem; Raphael Gottardo
• Keywords: Single-Cell Analysis; Gene regulatory network

Advances in high-throughput, single cell gene expression are allowing interrogation of heterogeneity. However, there is concern that the cycle phase a might bias characterizations at single-cell level. We assess effect on cells by measuring 333 genes 930 across three phases and lines. determine each cell's non-invasively without chemical arrest use it as covariate tests differential expression. observe bi-modal expression, previously-described phenomenon, wherein otherwise abundant either strongly positive, or undetectable within individual cells. This bi-modality likely both biologically technically driven. Irrespective its source, we show should be modeled to draw accurate inferences from experiments. To this end, propose semi-continuous modeling framework based generalized linear model, characterize with consistent effects Our new computational improves detection previously characterized cell-cycle compared approaches do not account for data. our modelling estimate co-expression networks. These networks suggest addition having phase-dependent shifts (when averaged over many cells), some, but all, canonical tend co-expressed groups amount variability attributable cycle. find explains only 5%–17% variability, suggesting will large nuisance factor analysis transcriptome.