JADE: Joint Alignment and Deep Embedding for Multi-Slice Spatial Transcriptomics.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Spatially resolved transcriptomics (SRT) generates gene expression data with spatial information.
• Analyzing multiple SRT slices is crucial for reconstructing tissue structures and identifying consistent biological patterns.
• Existing methods struggle with physical distortions, technical variability, and batch effects in multi-slice SRT analysis.

Purpose of the Study:

• To develop a unified computational framework for joint analysis of multi-slice SRT data.
• To simultaneously address spatial alignment and feature integration across tissue slices.
• To overcome challenges posed by distortions, variability, and batch effects in multi-slice SRT.

Main Methods:

• Introduced Joint Alignment and Deep Embedding for multi-slice SRT (JADE), a unified framework.
• Employed a roundtrip framework alternating between alignment and embedding refinement.
• Utilized attention mechanisms for dynamic weighting of embedding dimensions to infer alignment.

Main Results:

• JADE jointly optimizes alignment and representation learning in a shared latent space for robust multi-slice integration.
• Demonstrated superior performance of JADE over existing methods on human DLPFC and axolotl brain datasets.
• Achieved accurate and scalable cross-slice analysis by bridging spatial alignment and feature integration.

Conclusions:

• JADE offers a novel and effective solution for multi-slice SRT data integration.
• The framework enables more accurate reconstruction of tissue structures and identification of conserved spatial gene expression patterns.
• JADE advances the field by providing a scalable and robust tool for complex spatial transcriptomics analyses.