An interpretable Bayesian clustering approach with feature selection for analyzing spatially resolved transcriptomics...

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Spatially resolved transcriptomics (SRT) offers high-resolution molecular profiling within tissues.
• Existing clustering methods for SRT data often lack interpretability due to reliance on ad hoc dimensionality reduction.
• Understanding tissue organization requires accurate partitioning into distinct spatial domains.

Purpose of the Study:

• To develop an interpretable and accurate method for clustering spots or cells in SRT data.
• To integrate both molecular profiles and spatial information for improved domain identification.
• To identify key genes that define spatial domains.

Main Methods:

• A zero-inflated negative binomial mixture model for clustering based on molecular profiles.
• A feature selection mechanism to identify discriminating genes for interpretability.
• Incorporation of spatial information using a Markov random field prior.

Main Results:

• The proposed model demonstrates improved clustering accuracy compared to existing methods.
• The feature selection identifies biologically relevant genes that characterize spatial domains.
• Successful application to three real-world SRT datasets.

Conclusions:

• The joint modeling strategy effectively clusters SRT data by integrating molecular and spatial information.
• The method enhances interpretability through gene feature selection.
• This approach advances the analysis of tissue architecture using SRT data.