Spatiotemporal cell type deconvolution leveraging tissue structure

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Spot-based spatial transcriptomics (ST) provides aggregated transcriptomic data from tissue locations.
• Current deconvolution methods often rely on 2D spatial information and aggregated cell type references, limiting accuracy in heterogeneous tissues.
• Existing methods do not fully utilize multi-slice ST data or capture cell state variations.

Purpose of the Study:

• To develop a novel deconvolution method, SpaDecoder, that effectively utilizes 3D tissue structure and individual single-cell RNA sequencing (scRNA-seq) profiles.
• To improve the accuracy of cell type proportion estimation in spatial transcriptomics data, particularly in complex and heterogeneous tissues.
• To enable downstream analyses such as identifying cell type regions, colocalized cell types, and predicting 3D cell locations.

Main Methods:

• SpaDecoder employs a parallelized, per-spot deconvolution approach using a matrix factorization objective.
• It incorporates slice alignment, per-spot spatio-transcriptomic neighborhood inference, and 3D spatial Gaussian kernel weights.
• The method models individual scRNA-seq profiles to capture cell state variability, moving beyond aggregated cell type references.

Main Results:

• SpaDecoder accurately predicts cell type proportions by effectively leveraging 3D tissue structure and individual cell profiles.
• The method demonstrates superior performance compared to existing deconvolution techniques across various datasets and scenarios.
• SpaDecoder enables insightful downstream analyses, including uncovering key cell type regions, identifying colocalized cell types, and predicting 3D cell locations.

Conclusions:

• SpaDecoder significantly enhances deconvolution accuracy in spatial transcriptomics by integrating 3D structural information and detailed single-cell data.
• The method's ability to model cell state variability and adapt to heterogeneous environments provides more interpretable biological insights.
• SpaDecoder offers a powerful tool for analyzing multi-slice spatial transcriptomics data and advancing our understanding of tissue organization and cell interactions.