POPARI: Modeling multisample variation in spatial transcriptomics

Area of Science:

• Computational Biology
• Genomics
• Spatial Transcriptomics

Background:

• Integrating spatially-resolved transcriptomics (SRT) across biological samples is crucial for understanding in situ tissue architecture and cell-cell interactions.
• Existing computational tools are limited for multisample SRT analysis, hindering the study of condition-specific spatial variations.

Purpose of the Study:

• To introduce Popari, a novel probabilistic graphical model for factor-based decomposition of multisample SRT data.
• To capture and analyze condition-specific changes in spatial organization and gene expression programs across multiple samples.

Main Methods:

• Popari jointly learns spatial metagenes (gene expression programs) and their spatial affinities across samples.
• Key innovations include a differential prior for spatial accordance regularization and spatial downsampling for multiresolution analysis.
• The model was validated through simulations and applied to real-world datasets (mouse brain, thymus, ovarian cancer).

Main Results:

• Popari outperforms existing methods in multisample and multi-resolution spatial metrics.
• The tool successfully identified spatial metagene dynamics, spatial accordance, and cell identities in diverse biological samples.
• Specific applications revealed AD-linked metagenes in mouse brain, V(D)J recombination and T cell proliferation in thymus, and malignant-immune interactions in ovarian cancer.

Conclusions:

• Popari offers a general and interpretable framework for analyzing variations in multisample spatially-resolved transcriptomics data.
• The method enhances the understanding of spatial organization and cell-cell interactions in complex biological systems.