DOT: a flexible multi-objective optimization framework for transferring features across single-cell and spatial...

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell transcriptomics offers high gene coverage but lacks spatial information.
• Spatially-resolved technologies provide location data but often have limited gene coverage or resolution.
• Integrating these complementary modalities is crucial for comprehensive biological insights.

Purpose of the Study:

• To introduce DOT, a multi-objective optimization framework for integrating single-cell transcriptomics with spatial data.
• To overcome the limitations of individual technologies by transferring cellular features across modalities.
• To enable accurate cell population mapping and gene expression inference in complex tissue environments.

Main Methods:

• DOT employs a multi-objective optimization approach to transfer cellular features.
• It leverages genes beyond common sets and exploits local spatial context.
• A fast implementation utilizes the Frank-Wolfe algorithm for efficient computation.

Main Results:

• DOT successfully bridges single-cell transcriptomics with high- and low-resolution spatial data.
• The framework accurately localizes cell features in spatial datasets.
• It demonstrates improved estimation of unmeasured gene expression in low-coverage spatial data.

Conclusions:

• DOT provides a robust framework for integrating diverse transcriptomic and spatial data.
• This integration enhances the understanding of cellular heterogeneity and tissue architecture.
• The method offers state-of-the-art performance in spatial cell feature localization and gene expression analysis.