PlantDeconv: a structure-guided probabilistic deconvolution framework for plant spatial transcriptomics

Area of Science:

• Plant biology
• Genomics
• Bioinformatics

Background:

• Plant tissues exhibit unique spatial organization due to immobile cells and distinct anatomical structures.
• Existing spatial deconvolution methods often overlook plant-specific tissue architecture, limiting biological interpretability.
• Accurate interpretation of spatial transcriptomic data in plants requires methods that account for tissue structure.

Purpose of the Study:

• To develop a novel structure-guided probabilistic deconvolution framework for plant spatial transcriptomics.
• To improve the anatomical consistency and biological interpretability of spatial deconvolution in plant tissues.
• To address limitations of current methods that do not explicitly encode plant tissue structure.

Main Methods:

• Introduced PlantDeconv, a framework linking single-cell expression profiles to spatial transcriptomic spots.
• Incorporated four plant-specific constraints: cell-type-to-region prior, expression-aware spatial continuity, differentiation-gradient constraint, and adaptive regularization.
• Validated using multi-resolution pseudo-spot benchmarks and two real plant datasets.

Main Results:

• PlantDeconv demonstrated low composition error in pseudo-spot benchmarks.
• Achieved enhanced regional enrichment and clearer anatomical boundaries in real plant datasets.
• Reduced ectopic spread, particularly in complex or continuously developing plant tissues.

Conclusions:

• Plant tissue structure can be leveraged to significantly improve spatial deconvolution.
• PlantDeconv offers a more anatomically consistent and biologically interpretable approach for plant spatial transcriptomics.
• This framework advances the analysis of gene expression patterns within plant anatomical contexts.