Recovering Gene Interactions from Single-Cell Data Using Data Diffusion

Abstract

Single-cell RNA-sequencing is revolutionizing biological discovery. However, scRNA-seq technologies suffer from many sources of significant technical noise, the most prominent being undersampling of mRNA molecules, often termed ‘dropout’. Dropout can severely obscure important gene-gene relationships and impedes the possibility of learning gene regulatory networks at single cell resolution. To address this, we developed MAGIC (Markov Affinity-based Graph Imputation of Cells), a computational approach that shares information across similar cells, via data diffusion, to correct the mRNA count matrix and fill in missing transcripts. We validate MAGIC on a number of biological systems and find it effective at recovering gene-gene relationships and additional structures. We use MAGIC to explore the epithelial-to-mesenchymal transition (EMT) and reveal a phenotypic continuum of states, with the majority of cells residing in intermediate states that display stem-like signatures. Further, MAGIC uncovers the dynamics of gene expression underlying EMT, including known and novel regulatory interactions, demonstrating that our approach is able to successfully predict regulatory relations without perturbations.