spatiAlign: an unsupervised contrastive learning model for data integration of spatially resolved transcriptomics.

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• Integrative analysis of spatially resolved transcriptomics data is crucial for understanding complex biological systems.
• Integrating multiple tissue sections faces challenges in batch effect removal, especially with diverse technologies or collection times.

Purpose of the Study:

• To introduce spatiAlign, an unsupervised contrastive learning model for integrating multiple tissue sections.
• To enable joint downstream analysis of multiple spatially resolved transcriptomics datasets.

Main Methods:

• spatiAlign utilizes gene expression and cell spatial location for integration.
• The model performs unsupervised contrastive learning.
• It enables analysis in low-dimensional embeddings and the reconstructed full expression space.

Main Results:

• spatiAlign outperforms state-of-the-art methods in learning joint and discriminative representations.
• The model effectively handles complex batch effects and distinct biological characteristics across tissue sections.
• Demonstrated benefits in integrative analysis of time-series brain sections.

Conclusions:

• spatiAlign provides a robust solution for integrating multiple tissue sections in spatially resolved transcriptomics.
• The method enhances downstream analyses like spatial clustering, differential expression, and trajectory inference.
• spatiAlign facilitates a deeper understanding of biological systems through integrated spatial data analysis.