The proliferation of single cell datasets has brought a wealth of information, but also great challenges in data analysis. Obtaining a cohesive overview of multiple single cell samples is difficult and requires consideration of cell population structure - which may or may not be well defined - along with subtle shifts in expression within cell populations across samples, and changes in population frequency across samples. Ideally, all this would be integrated with the experimental design, e.g. time point, genotype, treatment etc. Data visualisation is the most effective way of communicating analysis but often this takes the form of a plethora of t-SNE plots, colour coded according to marker and sample. In this manuscript, I introduce a novel exploratory data analysis and visualisation method that is centred around a novel quasi-distance (DensityMorph) between single cell samples. DensityMorph makes it possible to plot single cell samples in a manner analogous to performing principal component analysis on microarray samples. Biological interpretation is ensured by the introduction of Explanatory Components, which show how marker expression and coexpression drive the differences between samples. This method is a breakthrough in terms of displaying the most pertinent biological changes across single cell samples in a compact plot. Finally, it can be used either as a stand-alone method or to structure other types of analysis such as manual flow cytometry gating or cell population clustering.
Comparing single cell datasets using DensityMorph
