Reproduce “SIGMMA: Hierarchical Graph-Based Multi-Scale Multi-modal Contrastive Alignment of Histopathology Image and Spatial Transcriptome” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Dabin Jeong, Amirhossein Vahidi, Ciro Ramírez-Suástegui, Marie Moullet, Kevin Ly, Mohammad Vali Sanian, Sebastian Birk, Yinshui Chang and 6 more
Researchers at the Wellcome Sanger Institute developed SIGMMA, a hierarchical graph-based framework that integrates histopathology images and spatial transcriptomics data across multiple scales. This method achieved superior performance in gene expression prediction and cross-modal retrieval, while also revealing biologically meaningful tissue microenvironments.
Key insights
- Explicitly modeling and aligning histopathology images and spatial transcriptomics data across hierarchical scales (micro, meso, macro) provides a more comprehensive and biologically accurate view of tissue architecture.
- Representing spatial transcriptomics data as cell graphs, rather than 1D sequences, is critical for preserving spatial topology and cell-cell interactions, which are fundamental to tissue function.
- A novel graph learning module with stochastic edge addition constrained by neighbor-patches effectively reconciles the differing receptive fields between image regions and graph connectivity, enabling coherent multi-scale cross-modal correspondence.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.