Unveiling spatial domains from spatial multi-omics data using dual-graph regularized ensemble learning.

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• Spatial multi-omics technologies enable simultaneous analysis of molecular profiles and spatial locations, offering insights into cellular heterogeneity and tissue microenvironments.
• Data sparsity and diverse distributions pose challenges for integrating and analyzing spatial multi-omics data.

Purpose of the Study:

• To propose SMODEL, a novel ensemble learning framework for detecting spatial domains from spatial multi-omics data.
• To address challenges in spatial multi-omics data integration and analysis.

Main Methods:

• SMODEL utilizes an ensemble learning framework based on dual-graph regularized anchor concept factorization.
• It employs an element-wise weighted ensemble strategy to integrate multiple base clustering results.
• Anchor concept factorization and dual-graph regularization are used to learn robust spatial consensus representations.

Main Results:

• SMODEL was evaluated on real and simulated spatial multi-omics datasets across various technologies, tissue types, and species.
• Experimental results show SMODEL outperforms existing methods in spatial domain identification.
• The framework effectively captures tissue structure, enhancing the understanding of cellular heterogeneity.

Conclusions:

• SMODEL provides a robust and effective approach for spatial domain detection in multi-omics data.
• The method enhances the interpretation of tissue architecture and cellular composition.
• SMODEL contributes to advancing the analysis of complex biological systems using spatial multi-omics data.