MaskGraphene: an advanced framework for interpretable joint representation for multi-slice, multi-condition spatial...

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• Spatial transcriptomics (ST) enables gene expression analysis within tissue context.
• Integrating multiple ST slices is crucial for comprehensive biological understanding.
• Batch effects and spatial geometry preservation pose significant challenges in ST data integration.

Purpose of the Study:

• To develop a novel method for joint analysis of multiple spatial transcriptomics slices.
• To generate interpretable embeddings that preserve spatial information and correct for batch effects.
• To improve downstream analysis applications in spatial transcriptomics.

Main Methods:

• MaskGraphene, a graph neural network approach utilizing masked self-supervised learning.
• Integration of triplet loss and cluster-wise local alignment for enhanced embedding.
• Establishment of indirect "soft-links" and direct "hard-links" between ST slices.

Main Results:

• MaskGraphene achieved superior alignment and interpretability compared to eight benchmark methods.
• The method generated joint embeddings with high geometric fidelity across multiple slices.
• Demonstrated enhanced performance in downstream tasks like domain identification and biomarker discovery.

Conclusions:

• MaskGraphene provides a robust framework for integrating diverse spatial transcriptomics datasets.
• The approach facilitates deeper biological insights through improved data alignment and interpretability.
• Enables advanced applications in ST, including trajectory reconstruction and brain-layer mapping.