Large-scale GNN-based cell segmentation for spatial transcriptomics. Scaling to 10M+ nodes/30M transcripts with 1000x speedup via multi-GPU distributed training.

Supervised graph-attention framework for mapping single-cell gene expression to spatial locations. Outperforms Tangram/NovoSpaRc.

Geometric deep learning framework on single-cell gene regulatory networks for cell annotation and discovery.

Structured dependency modeling and visualization for mixed-type multivariate data using probabilistic graphical models.