SIMVI reveals intrinsic and spatial-induced states in spatial omics data

Area of Science:

• Genomics
• Computational Biology
• Systems Biology

Background:

• Spatial omics technologies provide gene expression data within tissue context.
• Current computational methods struggle to differentiate intrinsic cellular variability from intercellular spatial interactions.
• This limitation hinders the accurate modeling of spatial gene regulation and biological discovery.

Purpose of the Study:

• To introduce Spatial Interaction Modeling using Variational Inference (SIMVI), an annotation-free computational framework.
• To disentangle cell-intrinsic and spatial-induced latent variables in spatial omics data.
• To enable accurate estimation of single-cell resolution spatial effects (SE) for downstream biological analysis.

Main Methods:

• Developed SIMVI, a variational inference-based framework for spatial omics data analysis.
• Provided theoretical support for disentangling intrinsic and spatial variations.
• Applied SIMVI to diverse spatial omics datasets including MERFISH, Slide-seqv2, Slide-tags, spatial multiome, and CosMx melanoma.

Main Results:

• SIMVI effectively disentangles cellular variations and infers accurate spatial effects across multiple platforms and tissues.
• The framework uniquely uncovers significant spatial regulations and biological dynamics.
• SIMVI revealed cyclical spatial dynamics of germinal center B cells in tonsil data and identified space-and-outcome-dependent macrophage states in melanoma.

Conclusions:

• SIMVI offers a robust method for modeling gene expression in spatial omics, accurately distinguishing intrinsic and spatial factors.
• The framework facilitates novel downstream analyses, uncovering complex biological insights in various tissues.
• SIMVI demonstrates significant potential for advancing our understanding of cellular interactions and tissue organization in health and disease.