Latent factor modelling of scRNA-seq data uncovers novel pathways dysregulated in cell subsets of autoimmune disease patients

Abstract

Latent factor modelling applied to single-cell RNA-sequencing (scRNA-seq) data is a useful approach to discover gene signatures associated with cell states. However, it is often unclear what method is best suited for specific tasks and how latent factors should be interpreted from a biological perspective. Here, we compare four state-of-the-art methods and explore their stability, predictive power and coverage of known biology. We then propose an approach that leverages the derived latent factors to directly assign pathway activities to specific cell subsets. By applying this framework to scRNA-seq datasets from biopsies of rheumatoid arthritis and systemic lupus erythematosus patients, we discover both known and novel disease-relevant gene signatures in specific cellular subsets in a fully unsupervised way. Focusing on rheumatoid arthritis, we identify an inflammatory Oncostatin M receptor signalling signature active in a subset of synovial fibroblasts and dysregulation of the GAS6 - MERTK axis in a subset of synovial monocytes with efferocytic function. Overall, we provide insights into strengths and weaknesses of latent factors models for the analysis of scRNA-seq data, we develop a framework to identify cell subtypes in a function- or phenotype-driven way and use it to identify novel pathways dysregulated in rheumatoid arthritis.