Compound-SNE: Comparative alignment of t-SNEs for multiple single-cell omics data visualisation.

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• Single-cell omics data analysis relies on visualization for cell-type separation.
• Data integration and batch correction methods merge multiple datasets for downstream analysis.
• Existing methods alter feature spaces, obscuring sample-specific features and local structures.

Purpose of the Study:

• To introduce Compound-SNE, a novel method for enhanced visualization of single-cell omics data.
• To enable effective visual comparisons across numerous samples, including different patients, omic modalities, or time points.
• To address limitations of current data integration methods that obscure sample-specific features.

Main Methods:

• Compound-SNE performs a 'soft alignment' of samples in embedding space.
• This approach aligns cell types across different datasets.
• Preserves local embedding structures present in independently embedded samples.

Main Results:

• Compound-SNE successfully aligns cell types in embedding space across multiple samples.
• The method retains local embedding structures that are typically lost during batch correction.
• Enables improved visual comparisons of complex, multi-sample omics datasets.

Conclusions:

• Compound-SNE enhances the visualization of single-cell omics data by balancing data integration and sample-specific feature preservation.
• It provides a valuable tool for researchers analyzing large, heterogeneous single-cell datasets.
• The method facilitates clearer interpretation of cell-type relationships and biological variations across samples.