High-parameter spatial multi-omics through histology-anchored integration

Area of Science:

• Computational biology
• Genomics
• Histopathology

Background:

• Spatial omics technologies aim for high-parameter, multi-omics coprofiling but face integration challenges.
• Serial-section profiling introduces the spatial diagonal integration problem when combining complementary panels.

Purpose of the Study:

• To develop computational frameworks (SpatialEx and SpatialEx+) for integrating spatial molecular data across tissue sections.
• To leverage histology as a universal anchor for multi-omics data integration in spatial profiling.

Main Methods:

• SpatialEx utilizes a pretrained hematoxylin and eosin foundation model with hypergraph and contrastive learning to predict single-cell omics from histology.
• SpatialEx+ incorporates an omics cycle module for cross-omics consistency via slice-invariant mappings, enabling integration without comeasured data.
• The frameworks encode multi-neighborhood spatial dependencies and global tissue context.

Main Results:

• Demonstrated superior hematoxylin and eosin-to-omics prediction and diagonal integration of panels and omics.
• Validated across various biological scenarios, showing robustness with nonoverlapping or heterogeneous sections.
• Frameworks scale to over 1 million cells and support unlimited omics layers.

Conclusions:

• SpatialEx and SpatialEx+ provide a broadly accessible solution for multimodal spatial profiling.
• Histology-guided integration overcomes key challenges in serial-section spatial omics.
• The developed frameworks facilitate seamless and accurate multi-omics data integration.