Enhancing spatial domain detection in spatial transcriptomics with EnSDD

Area of Science:

• Genomics and Computational Biology
• Spatial Transcriptomics
• Bioinformatics

Background:

• Spatial transcriptomics revolutionizes organ function and tissue microenvironment studies.
• Accurate identification of spatial domains for genome heterogeneity and cellular interactions is challenging.

Purpose of the Study:

• To introduce EnSDD (Ensemble-learning for Spatial Domain Detection), a novel method for automated spatial domain identification.
• To enhance the analysis of tissue heterogeneity and cellular interactions in spatial transcriptomics data.

Main Methods:

• EnSDD integrates eight state-of-the-art spatial domain detection methods using ensemble learning.
• A dynamic weighting mechanism optimizes base model contributions and provides performance evaluation without ground truth.
• Identifies domain-specific spatially variable genes and cell type distributions.

Main Results:

• EnSDD significantly improves the accuracy of spatial domain identification across diverse datasets.
• The method successfully detects genes with distinct spatial expression patterns.
• Reveals domain-specific cell type enrichment, offering deeper insights into tissue regionalization.

Conclusions:

• EnSDD provides a robust and accurate approach for spatial domain detection in transcriptomics.
• The method offers valuable insights into tissue spatial heterogeneity and cellular organization.
• EnSDD advances the analytical capabilities for understanding complex biological tissues.