Cells interact with each other for proper function and homeostasis. Often, co-expression of ligand-receptor pairs from the single-cell RNAseq (scRNAseq) has been used to identify interacting cell types. Recently, RNA sequencing of physically interacting multi-cells has been used to identify interacting cell types without relying on co-expression of ligand-receptor pairs. This opens a new avenue to study the expression of interacting cell types. We present DeepDoublet, a deep-learning-based tool to decompose the transcriptome of physically interacting two cells (or doublet) into two sets of transcriptome. Applying DeepDoublet to the doublets of hepatocyte and liver endothelial cells (LECs), we successfully decomposed into the transcriptome of each cell type. Especially, DeepDoublet identified specific expression of hepatocytes when they are interacting with LECs. Among them was Angptl3 which has a role in blood vessel formation. DeepDoublet is a tool to identify neighboring cell-dependent gene expression.
Read full abstract