STIFT: spatiotemporal transcriptomics integration through spatially informed multi-timepoint bridging.

Area of Science:

• Computational Biology
• Developmental Biology
• Regenerative Medicine

Background:

• Spatial transcriptomics provides gene expression data with spatial resolution.
• Integrating spatial and temporal dimensions is crucial for understanding dynamic biological processes.
• Existing methods struggle to effectively integrate large-scale spatiotemporal transcriptomics data.

Purpose of the Study:

• To present STIFT (SpatioTemporal Integration Framework for Transcriptomics), a novel computational framework.
• To enable the integration and analysis of large-scale 2D or 3D spatiotemporal transcriptomics data.
• To facilitate the exploration of developmental dynamics and regenerative processes.

Main Methods:

• STIFT combines developmental spatiotemporal optimal transport, spatiotemporal graph construction, and a graph attention autoencoder.
• The framework utilizes temporal triplet learning for enhanced analysis.
• It supports integration of large-scale datasets, including batch effect removal.

Main Results:

• STIFT successfully integrated and analyzed complex spatiotemporal transcriptomics datasets from axolotl brain regeneration, mouse embryonic development, and planarian regeneration.
• The framework effectively removed batch effects and identified distinct spatial domains.
• Temporal developmental patterns and biological variations were preserved across hundreds of thousands of data points.

Conclusions:

• STIFT is an effective and specific framework for integrating spatiotemporal transcriptomics data.
• It enables comprehensive analysis of spatial domains, developmental trajectories, and biological variations.
• The framework advances the study of complex biological processes across space and time.