Integration of spatial and single-cell data across modalities with weakly linked features.

Area of Science:

• Multi-omics data integration
• Computational biology
• Single-cell analysis

Background:

• Current multi-modal single-cell technologies cannot capture all biological information within a single cell.
• Existing data integration methods require highly correlated features for successful cross-modal analysis.
• Weakly linked features pose a significant challenge for integrating diverse biological datasets.

Purpose of the Study:

• To develop a novel cross-modal data integration method, MaxFuse, capable of high-quality integration even with weakly linked features.
• To demonstrate the robustness and accuracy of MaxFuse across various benchmarking datasets.
• To enable the spatial consolidation of multi-omic information at single-cell resolution.

Main Methods:

• MaxFuse utilizes iterative co-embedding, data smoothing, and cell matching for cross-modal integration.
• The method is modality-agnostic, allowing flexibility in combining different data types.
• It leverages all available information within each modality to enhance integration quality.

Main Results:

• MaxFuse achieves significant improvements (20-70%) over existing methods in weak linkage scenarios.
• Demonstrates high robustness and accuracy on benchmarking datasets.
• Successfully integrated spatial proteomic, transcriptomic, and epigenomic data in tissue sections at single-cell resolution.

Conclusions:

• MaxFuse overcomes limitations of current methods by effectively integrating data with weakly linked features.
• Enables comprehensive spatial multi-omic analysis, advancing biological insights.
• Offers a powerful tool for researchers working with complex single-cell and spatial omics data.