Prof. Dr. Stein Aerts
Laboratory of Computational Biology, KU Leuven, Belgium
"From single-cell multi-omics to gene regulatory networks and enhancer logic"
Single-cell transcriptomics and single-cell epigenomics allow building cell atlases of any tissue and species, providing new opportunities to predict gene regulatory networks that control the identity of cell types and cell states.
Here I will present a new computational strategy that we developed, that takes advantage of the joint analysis of scRNA-seq and scATAC-seq data, and that derives “enhancer-GRNs” (eGRN) with key transcription factors, genomic enhancers, and predicted target genes per cell type. On the scATAC-seq side, our strategy exploits topic modelling and motif discovery in co-accessible regions to predict TFs. On the scRNA-seq side, we use random forest regression (GENIE3) to link both accessible regions, as well as upstream TFs, to candidate target genes. In parallel, we use deep learning on the scATAC-seq topics to prioritize enhancers based on TF motif combinations. I will discuss the results of two case studies where we applied this strategy to, namely the Drosophila brain and the mouse liver. For the fly brain, we profiled 240K cells with scATAC-seq and 118K cells with scRNA-seq, and found eGRNs for 40 cell types. The eGRNs and deep learning predictions of enhancers were validated with in vivo enhancer reporter assays. For the mouse liver, we performed scATAC-seq, scRNA-seq, sc-multi-ome, and spatial omics, and predict eGRNs for the different liver cell types, and for zonated hepatocyte states. These case studies illustrate recent advances, as well as current limitations, in the field of single-cell regulatory genomics.