Toward Computationally Complete Spatial Omics.

Area of Science:

• Computational Biology
• Genomics
• Proteomics
• Metabolomics
• Histology

Background:

• Multimodal spatial omics technologies map molecular data in tissues but face limitations in simultaneous modality measurement and data quality.
• Current methods struggle to integrate diverse omics data, hindering comprehensive analysis of tissue complexity.

Purpose of the Study:

• To develop COSIE, a computational framework for generating high-resolution, multilayered molecular landscapes from spatial omics data.
• To integrate diverse omics modalities including epigenome, transcriptome, proteome, and metabolome with histology into a unified spatial representation.
• To enhance the analysis of tissue structure, data quality, and predictive capabilities across various spatial omics platforms.

Main Methods:

• Developed the COSIE computational framework to unify multimodal spatial omics data.
• Applied COSIE to 12 datasets across 10 spatial technologies, 8 modalities, and 9 tissue types.
• Validated COSIE's performance against existing methods in resolving tissue structures and enhancing data quality.

Main Results:

• COSIE successfully integrated histology, epigenome, transcriptome, proteome, and metabolome data into a unified representation.
• The framework demonstrated superior performance in resolving tissue structures, enhancing noisy measurements, and predicting unmeasured modalities.
• COSIE identified invasive tumor subregions associated with clinical outcomes and predicted spatial gene expression from histology images.

Conclusions:

• COSIE advances computationally complete spatial omics by transforming fragmented data into comprehensive spatial maps.
• The framework enables the creation of digital tissue twins for advanced biomedical research and clinical applications.
• COSIE's ability to integrate and enhance multimodal spatial omics data opens new avenues for biological discovery and precision medicine.