Using topic modeling to detect cellular crosstalk in scRNA-seq.

Area of Science:

• Single-cell genomics
• Computational biology
• Immunology

Background:

• Cell-cell interactions are crucial for biological processes.
• Current scRNA-seq interaction methods rely on curated ligand-receptor databases, limiting discovery.
• New protocols capture physically interacting cells, enabling knowledge-independent interaction studies.

Purpose of the Study:

• To introduce a novel computational method for detecting gene expression changes in interacting cells.
• To enable the study of cell-cell interactions without prior biological knowledge or curated databases.
• To identify novel genes and pathways involved in cell-cell communication.

Main Methods:

• Latent Dirichlet Allocation (LDA) model applied to scRNA-seq data.
• Analysis of synthetic datasets to validate gene detection accuracy.
• Application to datasets of physically interacting cells and standard droplet-based scRNA-seq data.

Main Results:

• Successfully identified genes changing due to cell-cell interaction in synthetic and real datasets.
• Highlighted adhesion and co-stimulatory molecules as indicators of physical interaction.
• Generated ranked gene lists for interacting cell subpopulations, revealing novel interaction candidates.

Conclusions:

• The developed LDA-based method effectively detects gene expression changes in interacting cells.
• This approach offers a powerful, knowledge-independent alternative for studying cell-cell interactions.
• The method is applicable to various scRNA-seq protocols, including those not specifically designed for capturing interacting cells.