Automatic identification of relevant genes from low-dimensional embeddings of single cell RNAseq data

Abstract

Dimensionality reduction is a key step in the analysis of single-cell RNA sequencing data and produces a low-dimensional embedding for visualization and as a calculation base for downstream analysis. Nonlinear techniques are most suitable to handle the intrinsic complexity of large, heterogeneous single cell data. With no linear relation between genes and embedding however, there is no way to extract the identity of genes most relevant for any cells position in the low-dimensional embedding, and thus the underlying process. In this paper, we introduce the concepts of global and local gene relevance to compute an equivalent of principal component analysis loadings for non-linear low-dimensional embeddings. While global gene relevance identifies drivers of the overall embedding, local gene relevance singles out genes that change in small, possibly rare subsets of cells. We apply our method to single-cell RNAseq datasets from different experimental protocols and to different low dimensional embedding techniques, shows our methods versatility to identify key genes for a variety of biological processes. To ensure reproducibility and ease of use, our method is released as part of destiny 3.0, a popular R package for building diffusion maps from single-cell transcriptomic data. It is readily available through Bioconductor.