Joint modeling of cellular heterogeneity and condition effects with scPCA in single-cell RNA-seq

Single-cell RNA sequencing (scRNA-seq) in multi-condition experiments enables the systematic assessment of treatment effects. Analyzing scRNA-seq data relies on linear dimensionality reduction (DR) methods like principal component analysis (PCA). These methods decompose high-dimensional gene expression profiles into interpretable factor representations and prototypical expression patterns (components). However, integrating study covariates within linear DR frameworks remains a challenging task. We present scPCA, a flexible DR framework that jointly models cellular heterogeneity and conditioning variables, allowing it to recover an integrated factor representation and reveal transcriptional changes across conditions and components of the decomposition. We show that scPCA extracts an interpretable latent representation by analyzing unstimulated and IFNß-treated PBMCs and show its utility in mitigating batch effects. We examine age-related changes in rodent lung cell populations, uncovering a previously unreported surge in Ccl5 expression in T cells. We illustrate how scPCA may be employed to identify coordinated transcriptional changes across multiple time-points in depolarized visual cortex neurons. Finally, we show that scPCA elucidates transcriptional shifts in CRISPR-Cas9 chordin knockout zebrafish single-cell data despite large difference cell abundance across conditions. scPCA is a general method applicable beyond scRNA-seq to other high-dimensional datasets.