eMCI: An Explainable Multimodal Correlation Integration Model for Unveiling Spatial Transcriptomics and Intercellular...

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Current methods for integrating single-cell RNA sequencing (scRNA-seq) and spatial transcriptomics (ST) data are task-specific.
• Existing approaches often provide incomplete ST data analysis, missing complex relationships between spatial expression, cell specificity, and intercellular communication.

Purpose of the Study:

• To develop an explainable multimodal correlation integration model (eMCI) for comprehensive ST data analysis.
• To leverage deep neural networks for fusing scRNA-seq and ST data, enabling multiple integrated analysis tasks at the cellular level.

Main Methods:

• Developed eMCI, a deep neural network-based model for multimodal data integration.
• Utilized spot-cell correlations to fuse scRNA-seq and ST data.
• Employed an attribution algorithm to identify key spatial components and elucidate cell-type specificity and intercellular communication patterns.

Main Results:

• eMCI demonstrated superior or comparable accuracy in cell-type classification and deconvolution compared to state-of-the-art methods on simulated and real ST datasets.
• The model successfully identified key spatial components linked to cell types and elucidated spatial expression patterns underlying cell specificity and communication.
• Applied to cross-species datasets (zebrafish, soybean, human lung), eMCI accurately estimated cell composition and inferred cellular interactions within spatial and temporal contexts.

Conclusions:

• eMCI provides an integrative framework for resolving spatial transcriptomes using scRNA-seq data.
• The model elucidates intercellular signal transduction mechanisms across spatial domains without requiring prior biological references.
• eMCI facilitates the discovery of spatial expression patterns specific to cell types and cell-cell communication.