Professor, Department of Computer Science, Beijing Normal-Hong Kong Baptist University (BNBU), Zhuhai, Guangdong, China.

Biography:

Weifeng SU received his Ph.D. degrees in Computer Science and Engineering from Hong Kong University of Science & Technology in 2007. He is currently an Professor in the Department of Computer Science at Beijing Normal-Hong Kong Baptist University (BNBU) in Zhuhai, Guangdong, China. He has authored or co-authored many peer-reviewed publications in leading journals and conferences, including NeurIPS, IEEE TKDE, IEEE TNNLS, ACM TODS, ACM TWeb.

CASH3D: Color-aware 3D Multimodal Hypergraph Learning for Revealing Biologically Meaningful Tissue Organization

Three-dimensional (3D) multimodal biological data are pivotal for understanding spatial tissue function and disease progression. However, most computational methods still struggle to recover biologically meaningful patterns from spatial transcriptomics (ST) data. Combining ST with paired histological staining images provides complementary signals: ST captures molecular expression profiles, whereas histology reveals cellular composition and spatial organization. The central challenge is to fuse these modalities to characterize tissue layers and microenvironments in 3D. To this end, we propose CASH3D, a biologically driven multimodal 3D hypergraph model for integrating ST and histological data. CASH3D introduces a color-aware spatial adjacency construction that links spatially non-adjacent yet functionally similar regions. Built on hypergraphs constructed with foundation model embeddings, CASH3D captures higher-order interactions among spatial regions by integrating gene expression and morphological cues into a unified 3D spatial hypergraph that reflects biological structure beyond geometric proximity. Experiments on multiple datasets, including complex tumor and immune microenvironments, demonstrate that CASH3D recovers spatial structures that better align with tissue architecture and remain more stable across slices, enabling faithful identification of 3D biological organization. Code is publicly available at https://github.com/csybdwbs/CASH3D.