PlantCCC prioritizes context-specific candidate ligand-receptor communication patterns in plant spatial transcriptomics.

Intercellular communication supports plant development and environmental responses, but its analysis in plant tissues is complicated by cell walls, plasmodesmata, and local tissue architecture. Spatial proximity therefore does not necessarily indicate effective communication. Plant ligand-receptor (L-R) resources also contain expanded gene families, homology-derived mappings, and uneven levels of experimental support. We developed PlantCCC, a spatially aware graph-learning framework that uses a plant L-R database as a candidate search space and combines residual spatial expression enhancement, a directed heterogeneous candidate graph, expression-gated spatial weighting, spatially aware multi-head graph attention, and self-supervised contrastive learning to prioritize context-specific candidate edges. In a semi-synthetic benchmark, PlantCCC distinguished TRUE pairs containing an injected interaction component from CONFOUNDER pairs showing tissue co-localization alone, and remained comparatively robust under dropout perturbation. In poplar stem analyses based on a homology-derived Populus candidate L-R set, and in an independent Arabidopsis Visium HD analysis based on Arabidopsis PlantPhoneDB entries, PlantCCC prioritized candidate L-R axes that were consistent with tissue architecture, spatial expression patterns, and prior evidence for the corresponding signaling modules. PlantCCC provides an interpretable computational framework for prioritizing context-specific candidate cell-cell communication patterns in plant spatial transcriptomics.