Topological Data Analysis for Unsupervised Feature Selection in Large Scale Spatial Omics Data Sets

Area of Science:

• Computational Biology
• Genomics
• Topological Data Analysis

Background:

• Spatial transcriptomics generates large datasets requiring new analysis methods.
• Existing methods for comparing spatial gene expression patterns have limitations.

Purpose of the Study:

• To apply persistent homology for continuous quantification of spatial gene expression structure.
• To demonstrate its utility in identifying spatially variable genes and analyzing spatial omics data.

Main Methods:

• Utilized persistent homology, a topological data analysis technique.
• Applied the method to public spatial transcriptomics datasets (kidney disease, myocardial infarction).
• Extended the methodology to a spatial metabolomics sample.

Main Results:

• Developed a continuous measure of spatial gene expression structure.
• Successfully identified biologically meaningful insights in disease datasets.
• Demonstrated applicability across different spatial omics modalities.

Conclusions:

• Persistent homology offers advantages over p-value based methods for spatially variable gene identification.
• The approach facilitates unified analysis across diverse spatial omics data.
• Highlights the utility of topological data analysis in big data applications.