GAMMI: graph-guided contrastive and adversarial integration of single-cell and spatial multi-omics data

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• Integrating single-cell and spatial multi-omics data is crucial for understanding cellular functions and tissue organization.
• Realistic experimental designs often yield incomplete, unpaired, and overlapping data across batches and platforms, creating mosaic settings.
• Existing integration methods struggle with missing data, batch effects, and limited correspondence, especially when spatial structure is not modeled.

Purpose of the Study:

• To develop a unified graph learning framework for mosaic integration of heterogeneous single-cell and spatial multi-omics data.
• To address challenges posed by incomplete, unpaired, and partially overlapping multi-omics datasets.
• To improve the stability and accuracy of multi-omics data integration, particularly in low-overlap scenarios.

Main Methods:

• Introduced GAMMI (Graph-guided Adversarial Mosaic Multi-omics Integration), a novel graph learning framework.
• Employed heterogeneous graphs to jointly embed cells and molecular features, capturing cell-feature, feature-feature, and spatial relationships.
• Utilized an edge-based contrastive objective with missingness-aware negative sampling and adversarial domain adaptation for robust integration.
• Incorporated spatial data as structural constraints to enrich molecular representations.

Main Results:

• GAMMI effectively integrates mosaic single-cell and spatial multi-omics data, outperforming existing methods.
• Demonstrated superior performance in biological conservation, batch correction, and spatial reconstruction across diverse datasets.
• Showcased robustness in low-overlap and fully unpaired data regimes, addressing key limitations of current approaches.

Conclusions:

• GAMMI provides a powerful and unified framework for integrating complex, mosaic multi-omics data.
• The method enhances the resolution of cellular identity, regulatory programs, and tissue organization.
• GAMMI offers a significant advancement for multi-omics data integration in various biological research areas.