DeepDoublet identifies neighboring cell-dependent gene expression.

Area of Science:

• Cell biology
• Genomics
• Bioinformatics

Background:

• Cell-cell interactions are crucial for tissue function and homeostasis.
• Single-cell RNA sequencing (scRNAseq) traditionally infers interactions via ligand-receptor co-expression.
• Recent advancements allow RNA sequencing of physically interacting cells, enabling interaction studies without co-expression assumptions.

Purpose of the Study:

• To introduce DeepDoublet, a deep-learning tool for deconvoluting transcriptomes of physically interacting cell pairs (doublets).
• To enable the study of gene expression specific to interacting cell types.
• To identify novel cell-neighbor dependent gene expression patterns.

Main Methods:

• Development of DeepDoublet, a deep-learning algorithm designed to separate transcriptomic data from two physically interacting cells.
• Application of DeepDoublet to analyze doublets formed by hepatocytes and liver endothelial cells (LECs).
• Validation of DeepDoublet's ability to accurately decompose doublet transcriptomes into individual cell type profiles.

Main Results:

• DeepDoublet successfully decomposed transcriptomes from hepatocyte-LEC doublets into distinct hepatocyte and LEC profiles.
• The tool identified specific gene expression patterns in hepatocytes when interacting with LECs.
• Notably, Angiopoietin-like 3 (Angptl3), a gene involved in blood vessel formation, was identified as specifically expressed in interacting hepatocytes.

Conclusions:

• DeepDoublet provides a novel computational approach to analyze transcriptomes of physically interacting cells.
• The tool facilitates the discovery of cell-neighbor dependent gene expression.
• Findings highlight the potential of DeepDoublet in uncovering context-specific gene regulation during cell-cell interactions.