LETSmix: a spatially informed and learning-based domain adaptation method for cell-type deconvolution in spatial...

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Spatial transcriptomics (ST) offers insights into gene expression within tissue context.
• Limited resolution in current ST technologies results in mixed cell signals at each data point.
• Accurate cell type deconvolution is crucial for interpreting ST data.

Purpose of the Study:

• To develop a novel computational method, LETSmix, for accurate cell type deconvolution in ST data.
• To enhance the resolution and interpretability of ST data by addressing mixed-cell signals.
• To improve the integration of ST data with reference single-cell RNA sequencing (scRNA-seq) datasets.

Main Methods:

• LETSmix integrates spatial correlations using a tailored LETS filter, incorporating layer annotations, expression similarity, image texture, and spatial coordinates.
• A mixup-augmented domain adaptation strategy is employed to reconcile differences between ST and scRNA-seq data.
• The method refines ST data by leveraging multi-modal information for improved deconvolution.

Main Results:

• Comprehensive evaluations demonstrate LETSmix's high accuracy in estimating cell-type proportions across diverse ST platforms and tissue types.
• The method effectively deconvolves cell types, revealing accurate spatial patterns.
• LETSmix outperforms existing deconvolution methods in benchmark studies.

Conclusions:

• LETSmix provides a robust and accurate solution for cell type deconvolution in spatial transcriptomics.
• The method enhances the utility of ST data by improving spatial resolution and cell-type identification.
• LETSmix represents a significant advancement for spatial biology research and single-cell analysis.