Image-Enhanced Multi-Modal Contrastive Transformer for Subcellular Spatial Transcriptomics

Area of Science:

• Computational biology
• Molecular imaging
• Genomics

Background:

• Spatial molecular imaging offers high-resolution gene expression profiling but has limited gene detection.
• Integrating high-resolution imaging features with transcriptomic profiles is crucial for comprehensive subcellular analysis.

Purpose of the Study:

• To develop SIMMT, an image-enhanced multi-modal contrastive transformer framework.
• To identify spatial domains and enhance subcellular data by integrating morphology and transcriptomics.

Main Methods:

• A dual transformer architecture was designed to learn multi-modal cell representations from transcriptomics and morphology.
• A contrastive learning module was introduced to align tissue morphology and gene expression at the cell level.

Main Results:

• SIMMT consistently outperformed state-of-the-art methods in spatial clustering and gene expression pattern analysis across multiple datasets.
• The framework effectively identified tumor spatial heterogeneity and potential gene biomarkers in human lung and colorectal cancer tissues.

Conclusions:

• SIMMT provides a powerful framework for integrating spatial imaging and transcriptomic data for advanced subcellular analysis.
• This approach enhances the understanding of cellular heterogeneity and biomarker discovery in complex biological tissues.