Inferring biologically relevant molecular tissue substructures by agglomerative clustering of digitized spatial...

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Spatially resolved transcriptomics (SrT) enables the study of tissue architecture through gene expression patterns.
• Current computational methods often fail to fully utilize the spatial information inherent in SrT data.
• There is a need for advanced analytical tools to uncover the functional molecular substructures within complex tissues.

Purpose of the Study:

• To develop a novel computational method, MULTILAYER, for analyzing SrT data.
• To stratify SrT data into functionally relevant molecular substructures by exploiting spatial signatures.
• To enhance the understanding of tissue complexity through gene expression analysis.

Main Methods:

• Developed MULTILAYER, inspired by image analysis contextual pixel classification.
• Applied agglomerative clustering within locally defined transcriptomes (gexels).
• Utilized community detection methods for graphical partitioning of SrT data.

Main Results:

• MULTILAYER successfully resolves molecular tissue substructures across various SrT datasets.
• Demonstrated superior performance compared to common dimensionality reduction strategies.
• Achieved comparable performance to existing methods in detecting differentially expressed genes.
• Showcased capability to process high-resolution and multiple SrT datasets comparatively.

Conclusions:

• MULTILAYER offers a significant advancement in SrT data analysis by integrating spatial context.
• The method provides a digital image perspective, enabling contextual gexel classification.
• Opens avenues for developing self-supervised molecular diagnosis solutions.
• Addresses future needs for processing high-resolution and comparative SrT analyses.