scComm: a contrastive learning framework for deciphering cell-cell communications at single-cell resolution.

Area of Science:

• Computational Biology
• Single-cell Genomics
• Cancer Research

Background:

• Cell-cell communication is vital for multicellular organisms.
• Current single-cell RNA sequencing (scRNA-seq) methods often miss crucial details by averaging gene expression within cell clusters.
• This limitation hinders a comprehensive understanding of cellular interactions.

Purpose of the Study:

• To introduce scComm, a novel computational framework for high-resolution cell-cell communication inference.
• To overcome the limitations of existing scRNA-seq analysis methods.
• To reveal previously undetected biological insights from single-cell data.

Main Methods:

• Developed scComm, a computational framework utilizing supervised contrastive learning.
• Inferred cell-cell communications at the individual cell level, not just by clusters.
• Validated scComm's performance through simulations and application to real-world cancer datasets.

Main Results:

• scComm achieved up to 95% accuracy in simulations, outperforming existing methods.
• Analysis of colorectal cancer revealed communication links to PD-1 blockade response and tertiary lymphoid structures.
• In liver cancer, scComm identified three novel tumor subtypes and specific neutrophil subtypes involved in angiogenesis.

Conclusions:

• scComm enables unprecedented high-resolution analysis of cell-cell communication.
• The framework uncovers critical biological insights missed by traditional cluster-based approaches.
• scComm has significant potential for advancing cancer research and understanding tumor microenvironments.