SECTOR: structural entropy-based learning of spatiotemporal organisation in spatial transcriptomics.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Spatial transcriptomics (ST) enables gene expression analysis within tissue context.
• Existing methods struggle to jointly identify spatial domains and pseudotemporal trends.
• Current approaches may blur domain boundaries or lack integrated spatial awareness.

Purpose of the Study:

• To introduce SECTOR, a novel framework for unified spatial domain detection and pseudotime inference.
• To develop a lightweight deep graph learning approach for spatiotemporal modeling in ST.
• To improve clustering accuracy and pseudotime trend recovery in ST data.

Main Methods:

• SECTOR utilizes a deep graph learning framework with a differentiable structural entropy (SE) objective.
• It employs a fused spatial-expression graph and spatial total variation regularization.
• The method is evaluated on seven benchmark ST datasets.

Main Results:

• SECTOR consistently outperformed existing spatiotemporal methods in clustering accuracy.
• It matched or exceeded leading spatial clustering algorithms.
• Case studies demonstrated recovery of spatially organized pseudotime patterns.

Conclusions:

• SECTOR provides an effective and scalable strategy for modeling within-section spatiotemporal organization in ST.
• The structural entropy-based learning approach is robust for ST data analysis.
• SECTOR offers an integrated solution for spatial domain and pseudotime analysis.