CellContrast: Reconstructing spatial relationships in single-cell RNA sequencing data via deep contrastive learning.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) generates vast datasets but lacks spatial context, limiting analysis of complex biological activities.
• Reconstructing spatial relationships is crucial for understanding cellular interactions and tissue organization.

Purpose of the Study:

• To introduce CellContrast, a novel computational method for inferring spatial locations of single cells using spatial transcriptomics (ST) data.
• To enable accurate spatial reconstruction of scRNA-seq data by leveraging ST references.

Main Methods:

• CellContrast utilizes a contrastive learning framework trained on ST data.
• Gene expression data is projected into a hidden space where proximity is represented by similar values.
• Benchmarking performed on diverse ST platforms (SeqFISH, Stereo-seq, 10X Visium, MERSCOPE) using mouse embryo and human breast cell data.

Main Results:

• CellContrast significantly outperforms existing methods in spatial reconstruction accuracy for scRNA-seq data.
• Demonstrated effectiveness across multiple ST platforms and biological samples.
• Validated utility in cell-type co-localization and cell-cell communication analyses.

Conclusions:

• CellContrast provides accurate spatial mapping for scRNA-seq data, bridging the gap left by traditional methods.
• The recovered spatial information enhances biological discovery and mitigates false positives in downstream analyses.
• This method advances the integrated analysis of scRNA-seq and ST data for deeper biological insights.