LYNX: a deep generative model for linking spatial dynamics and cell interactions in multimodal spatial data

Area of Science:

• Spatial biology
• Computational biology
• Genomics and transcriptomics

Background:

• Tissues exhibit complex spatial organization with varying cell states and functions.
• Understanding spatial dynamics requires integrating multi-modal data but current methods have limitations.
• Existing tools struggle to model cell-cell communication across entire tissues.

Purpose of the Study:

• To introduce LYNX, a deep generative framework for unified spatial data analysis.
• To model how cell-cell interactions, phenotypes, and molecular programs vary along spatial gradients.
• To address limitations in existing computational approaches for multi-modal spatial data integration.

Main Methods:

• LYNX learns a shared latent representation from joint-measured modalities in 2D or 3D.
• It provides a unified coordinate system for analyzing spatial dynamics.
• The framework is compatible with diverse spatial profiling technologies and modalities.

Main Results:

• LYNX identifies novel spatial programs, such as metabolic interactions in liver and immune-tumor crosstalk in breast cancer.
• It robustly infers spatially resolved gradients and maps functional compartments.
• The framework successfully models cell-cell interactions along spatial axes.

Conclusions:

• LYNX offers a foundational and scalable framework for analyzing complex spatial biological data.
• It advances the understanding of healthy tissue physiology and disease progression.
• The tool enables decoding the temporal evolution of complex diseases through spatial analysis.