Quantitative characterization of cell niches in spatially resolved omics data

Area of Science:

• Computational biology
• Spatial omics
• Graph deep learning

Background:

• Spatial omics reveals colocalized cell communities (niches) coordinating tissue functions.
• Cellular interactions shape niches, but current computational methods rarely use this information.
• Existing methods lack quantitative characterization of niches based on communication pathways.

Purpose of the Study:

• Introduce NicheCompass, a novel graph deep-learning method for niche identification and characterization.
• Leverage cellular communication and signaling events to understand niche formation and function.
• Provide a scalable framework for spatial omics data analysis.

Main Methods:

• Developed NicheCompass, a graph deep-learning model utilizing cellular communication networks.
• Learned interpretable cell embeddings encoding signaling events for niche identification.
• Applied NicheCompass to mouse embryonic development and human cancer datasets.

Main Results:

• NicheCompass quantitatively characterizes niches based on communication pathways, outperforming existing methods.
• Successfully mapped tissue architecture in mouse development and delineated tumor niches in human cancers.
• Demonstrated scalability by constructing a large-scale spatial atlas of the mouse brain (8.4 million cells).

Conclusions:

• NicheCompass offers a scalable and effective framework for identifying and analyzing biological niches through cellular signaling.
• The method provides quantitative insights into niche composition and function.
• Enables cross-technology integration for spatial multi-omics data analysis.