SpatialCTD: A Large-Scale Tumor Microenvironment Spatial Transcriptomic Dataset to Evaluate Cell Type Deconvolution...

Area of Science:

• Genomics
• Computational Biology
• Immunology

Background:

• Spatially resolved transcriptomic profiling offers cost-effective multicellular resolution.
• Cell type deconvolution is crucial for analyzing mixed cell populations in spatial data.
• Existing benchmarks are limited, often simulated, mouse-based, and unsuitable for human immuno-oncology.

Purpose of the Study:

• Introduce SpatialCTD, a large-scale benchmark dataset for human immuno-oncology spatial transcriptomic deconvolution.
• Develop and validate GNNDeconvolver, a novel graph neural network method for cell type deconvolution.
• Provide an accessible tool for converting diverse spatial transcriptomic data into standardized pseudo spots.

Main Methods:

• Construction of SpatialCTD dataset using 1.8 million cells and 12,900 pseudo spots from human tumor microenvironments (lung, kidney, liver).
• Development of GNNDeconvolver, a graph neural network leveraging location-aware reference data.
• Comparative performance evaluation against state-of-the-art deconvolution methods.

Main Results:

• SpatialCTD offers a more realistic reference for deconvolution compared to single-cell RNA sequencing (scRNA-seq) derived references.
• GNNDeconvolver demonstrates superior performance over existing methods in cell type deconvolution.
• The proposed method achieves high accuracy without reliance on scRNA-seq data.

Conclusions:

• SpatialCTD and GNNDeconvolver provide a robust framework for advancing cell type deconvolution in human immuno-oncology.
• The developed tools facilitate comprehensive evaluation and analysis of spatial transcriptomic data across various platforms.
• This work addresses critical limitations in current benchmark datasets and deconvolution methodologies.