Network models for bridging denoising and identifying spatial domains of spatially resolved transcriptomics

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Spatially resolved transcriptomics (SRT) provides gene expression and spatial localization data for tissue architecture insights.
• Experimental procedures for SRT can introduce technical noise, impacting data quality.
• Current methods separate denoising and spatial domain identification, limiting performance.

Purpose of the Study:

• To develop an integrative network model, stACN, for joint denoising and spatial domain identification in SRT.
• To enhance the accuracy and applicability of spatial transcriptomics data analysis.
• To improve the understanding of tissue architecture through integrated data processing.

Main Methods:

• Proposed stACN (spatial transcriptomics Attribute Cell Network), an integrative network model.
• Employed a graph noise model to learn clean dual cell networks.
• Utilized joint tensor decomposition for deriving compatible cell features from denoised networks.

Main Results:

• stACN effectively denoises gene expression data in SRT.
• The model successfully identifies spatial domains within SRT datasets.
• Demonstrated enhanced data quality using Adjusted Rand Index (ARI) for clustering agreement.

Conclusions:

• stACN offers a unified approach for denoising and spatial domain analysis in SRT.
• The integrative model improves the reliability of spatial transcriptomics data.
• Facilitates more accurate insights into tissue architecture and cellular organization.