Gene regulatory network inference from single-cell RNA-seq reveals high-resolution cellular states

Cellular differentiation, from pluripotent stem cells to the diverse cell types in our body, is guided by complex gene regulatory networks (GRNs). However, determining the precise control associations that give rise to all the cell types and cellular states in the CNS is still a challenge. To explore landscapes of stable cellular states and their underlying regulatory networks, we have developed a new computational method: SCENIC (Single Cell rEgulatory Network Inference and Clustering). SCENIC allows to simultaneously reconstruct transcriptional regulatory networks and identify stable cell states using single-cell RNA-seq data. The gene networks are inferred based on co-expression and transcription factor binding site analysis. The active gene modules in each cell are subsequently used to identify the "attractor" states. We used SCENIC to re-analyze several publicly available datasets (e.g., the mouse cortex and hippocampus by Zeisel et al., Science 2015) and found that SCENIC accurately recovers the known cell types and infers meaningful GRNs with known (and novel) master regulators for each cell type without any prior knowledge of the system. Furthermore, clustering single cells based on the active gene modules rather than the raw expression profiles revealed cell states that with other approaches remain hidden, thereby compensating for batch effects (e.g. higher number of expressed genes in some cell types). In summary, our work provides an easy to use and flexible tool for the analysis of single-cell RNA-seq, and a statistically robust methodology that provides new insights into cell states based on the underlying gene regulatory network. Availability: SCENIC is implemented as an R/Bioconductor package.status: accepte