Microenvironment-aware spatial modeling for accurate inference of cell identity.

Area of Science:

• Spatial biology
• Computational biology
• Genomics

Background:

• Spatial omics technologies provide insights into cellular organization and tissue architecture at single-cell resolution.
• Existing computational methods for spatial omics primarily focus on domain detection, not cell identity inference.
• Traditional methods neglect the impact of the local microenvironment on cell states.

Purpose of the Study:

• To develop a computational method that integrates spatial context for improved cell identity modeling in spatial transcriptomics.
• To address the limitations of existing methods that rely solely on intrinsic molecular features.

Main Methods:

• Introduction of MEcell, a parameter-free computational method.
• MEcell explicitly incorporates and adaptively weights spatial context for cell identity inference.
• Validation across simulated and diverse real-world spatial transcriptomics datasets.

Main Results:

• MEcell accurately infers cell identities by leveraging spatial context.
• The method consistently outperformed existing approaches across multiple platforms (MERFISH/Vizgen, Xenium, CosMx, Visium HD, Slide-seqV2, open-ST) and tissue types.
• Demonstrated the critical role of the microenvironment in defining cell identity.

Conclusions:

• MEcell effectively captures spatially informed cellular heterogeneity.
• The findings underscore the importance of integrating spatial information for accurate cell type deconvolution.
• MEcell offers a powerful tool for analyzing complex spatial omics data.