Joint imputation and deconvolution of gene expression across spatial transcriptomics platforms.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Spatially resolved transcriptomics (SRT) technologies offer insights into gene expression within tissue microenvironments.
• Existing SRT platforms vary in spatial resolution, gene coverage, and sequencing depth.
• Integrating data from complementary SRT technologies can overcome individual platform limitations.

Purpose of the Study:

• To introduce Spatial Integration for Imputation and Deconvolution (SIID), a novel algorithm for integrating data from diverse SRT technologies.
• To enable accurate reconstruction of spatial gene expression matrices by leveraging paired observations.
• To address limitations in gene imputation and cell type deconvolution inherent in single SRT methods.

Main Methods:

• SIID employs spatial alignment to register data from different SRT modalities.
• A joint nonnegative factorization model is utilized for data integration and analysis.
• The algorithm reconstructs a latent spatial gene expression matrix from paired SRT data.

Main Results:

• Simulations demonstrate SIID's superior performance in spot-to-cell-type assignment and cell type-specific gene expression recovery compared to existing tools.
• SIID effectively imputes missing gene expression data when applied to paired SRT datasets.
• Application to real-world human breast and colon cancer data shows high accuracy in imputing holdout gene expression.

Conclusions:

• SIID provides a robust framework for integrating multimodal SRT data, enhancing spatial gene expression analysis.
• The algorithm overcomes limitations of individual SRT technologies, enabling more comprehensive tissue profiling.
• SIID has significant potential for advancing cancer research and other fields utilizing spatial transcriptomics.