Plant Spatio-Temporal Integration Network (PSTN): a framework for dynamic single-cell and spatial transcriptomics in fungal–plant interactions
TL;DR
The Plant Spatio-Temporal Integration Network (PSTN), which jointly learns stage-specific cell-to-space mappings through expression reconstruction and bidirectional maximum-similarity temporal regularization, enables temporally coupled single-cell and spatial integration in dynamic host–pathogen systems.
Abstract
Abstract Motivation Rice blast caused by Magnaporthe oryzae (M. oryzae) is a dynamic cross-kingdom infection process involving rapid transcriptional reprogramming and spatial tissue remodeling across successive stages. Although single-cell RNA sequencing (scRNA-seq) and spatial transcriptomics (ST) provide complementary information, most integration methods analyze stages independently and ignore temporal continuity. Results We developed the Plant Spatio-Temporal Integration Network (PSTN), which jointly learns stage-specific cell-to-space mappings through expression reconstruction and bidirectional maximum-similarity temporal regularization. Applied to matched rice scRNA-seq and ST data at 0, 12, and 24 h, PSTN outperformed Tangram, SpaGE, cell2location, and DestVI across three gene-set sizes. At 4000 highly variable genes (HVGs), mean squared error (MSE) remained below 0.09 at all stages, with high gene-wise and spot-wise correlations. Mapping-structure and marker-based analyses recovered tissue-associated spatial and temporal patterns. Gene-permutation controls markedly reduced performance, demonstrating reliance on correct cross-modal gene correspondence. An independent mouse-cortex benchmark supported the static reconstruction component across systems. PSTN enables temporally coupled single-cell and spatial integration in dynamic host–pathogen systems. Availability PSTN is available at https://github.com/zhuyuheng111/PSTN. The Zenodo DOI for the repository is 10.5281/zenodo.21365720.