STAID: A Self-Refining Deep Learning Framework for Spatial Cell-Type Deconvolution with Biologically Informed...

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Spatial transcriptomics enables gene expression analysis with spatial context.
• Accurate inference of cell-type composition within spatial transcriptomics data remains a challenge.

Purpose of the Study:

• To present STAID, a unified framework for accurate spot-level deconvolution of cell-type compositions in spatial transcriptomics data.
• To improve the understanding of tissue organization and cellular heterogeneity.

Main Methods:

• STAID integrates pseudo-spot generation with deep learning training via iterative refinement.
• It utilizes graph signal processing to capture higher-order gene-wise relationships.
• A self-reinforcing cycle enhances the accuracy of cell-type deconvolution.

Main Results:

• STAID outperforms existing methods in benchmarking studies.
• It accurately reconstructs cell-type spatial distributions and resolves cellular colocalization.
• Applied to clinical breast cancer, embryonic limb, and Crohn's disease datasets, STAID precisely infers distributions, associations, and organization.

Conclusions:

• STAID provides high-resolution cell-type distributions for spatial transcriptomics data.
• It offers deeper insights into tissue organization, cellular heterogeneity, and immune niches.
• The framework enhances tissue segmentation and reveals spatial associations crucial for biological understanding.