MultiSP deciphers tissue structure and multicellular communication from spatial multi-omics data.

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• Spatial multi-omics technologies offer simultaneous profiling of multiple data types while preserving tissue architecture.
• Interpreting complex spatial structures and cellular communication from sparse, noisy multi-omics data remains a significant challenge.

Purpose of the Study:

• To develop a deep learning framework, MultiSP, for enhanced representation and interpretation of spatial multi-omics data.
• To improve the understanding of tissue complexity, heterogeneity, and cell-cell communication within spatial contexts.

Main Methods:

• MultiSP employs efficient spatial and feature similarity fusion.
• It utilizes modality-specific probabilistic generative modeling and cross-modality adversarial learning.
• The framework integrates complementary information from multiple omics modalities.

Main Results:

• MultiSP outperforms existing methods in capturing biologically interpretable spatial domains.
• It effectively denoises spatial data, uncovers modality-specific spatial variations, and reveals gene regulation mechanisms.
• The framework identified fine-resolution cellular distribution maps in the tumor microenvironment, including prognosis-associated macrophage-enriched regions.

Conclusions:

• MultiSP provides a powerful framework for uncovering spatially multimodal heterogeneity and communication.
• The integration of complementary multi-omics data enhances the understanding of complex biological systems.
• This approach facilitates the inference of spatially resolved cell-cell communication networks.