Benchmarking cell type deconvolution in spatial transcriptomics and application to cancer immunotherapy.

Area of Science:

• Spatial transcriptomics
• Computational biology
• Immunotherapy research

Background:

• Accurate cell type deconvolution is crucial for spatial transcriptomics.
• Current methods' performance varies across biological contexts.
• Robust benchmarking is needed to assess deconvolution techniques.

Purpose of the Study:

• To introduce a realistic benchmarking framework for spatial deconvolution methods.
• To evaluate the performance of different deconvolution approaches.
• To investigate immune responses to anti-PD1 immunotherapy using spatial transcriptomics.

Main Methods:

• Developed a benchmarking framework with realistic simulations.
• Incorporated spatial and transcriptional complexity into simulations.
• Applied marker gene signature scoring and compared it with complex models.

Main Results:

• Marker gene scoring demonstrated competitive performance, outperforming complex models for rare cell types.
• The method successfully identified spatially localized immune and microenvironmental changes.
• Profound localized responses were observed in tumors and draining lymph nodes post-anti-PD1 therapy.

Conclusions:

• Marker gene signature scoring offers a robust and interpretable strategy for cell type deconvolution.
• The findings impact the interpretation of existing spatial transcriptomics studies.
• This approach is valuable for future analyses, especially without single-cell references.