Identifying multicellular spatiotemporal organization of cells with SpaceFlow

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Analyzing spatial transcriptomic data presents challenges in integrating cell expression similarity with spatial positioning.
• Existing methods often struggle to simultaneously capture both transcriptome similarity and spatial context.

Purpose of the Study:

• To introduce SpaceFlow, a novel deep learning framework for analyzing spatial transcriptomic data.
• To generate spatially-consistent low-dimensional embeddings that incorporate both expression similarity and spatial information.
• To develop a pseudo-Spatiotemporal Map for unraveling cellular spatiotemporal patterns.

Main Methods:

• Utilized spatially regularized deep graph networks to create embeddings.
• Integrated pseudotime concepts with cell spatial locations to form a pseudo-Spatiotemporal Map.
• Validated SpaceFlow on multiple spatial transcriptomic datasets at both spot and single-cell resolutions.

Main Results:

• SpaceFlow demonstrated robust domain segmentation and identified biologically meaningful spatiotemporal patterns.
• Successfully revealed evolving lineage in heart development data.
• Uncovered tumor-immune interactions in human breast cancer data.

Conclusions:

• SpaceFlow provides a flexible and powerful deep learning framework for spatial transcriptomic data analysis.
• The method effectively integrates spatiotemporal information, advancing the understanding of biological processes.
• SpaceFlow enhances the ability to identify complex cellular dynamics and interactions in situ.