MGCL-ST: multi-view graph contrastive learning for spatial transcriptomics imputation

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Spatial transcriptomics (ST) is crucial for analyzing tissue microenvironments.
• Current ST platforms face limitations in resolution and data completeness, creating spatial gaps.

Purpose of the Study:

• To develop a super-resolution imputation method for ST data.
• To enhance the accuracy and biological interpretability of spatial gene expression profiling.

Main Methods:

• Introduced MGCL-ST, a multi-view graph contrastive learning approach.
• Integrated local and global spatial graphs with histological features from a pathology foundation model.
• Applied the method to data from three diverse ST platforms.

Main Results:

• MGCL-ST demonstrated superior imputation accuracy compared to existing methods.
• Achieved enhanced spatial clustering and improved biological interpretability.
• Successfully mapped tumor microenvironments with greater precision.

Conclusions:

• MGCL-ST effectively addresses spatial gaps and resolution limitations in ST data.
• The method significantly advances the analysis of complex tissue architectures.
• Enables more precise investigations into tumor microenvironments and other biological systems.