1
0The paper introduces SCIGMA, a sophisticated deep learning framework designed to integrate diverse spatial multiomics data, such as transcriptomics, proteomics, and imaging. Traditional methods often struggle to maintain spatial context or scale to high-resolution datasets, but SCIGMA utilizes graph neural networks and an uncertainty-aware objective to produce clear biological insights. By quantifying model confidence at each location, the framework distinguishes true biological heterogeneity from technical noise across various tissue types and platforms. Evaluated on millions of spatial locations, it demonstrates superior reproducibility and accuracy in identifying anatomical structures compared to existing tools. Ultimately, SCIGMA offers a scalable and flexible solution for researchers seeking to map the complex molecular landscape of tissues with high precision.
References:
Chang S, Fleischmann A, Ma Y. Scalable, generalizable and uncertainty-aware integration of spatial multiomics across diverse modalities and platforms with SCIGMA[J]. Nature Genetics, 2026: 1-19.
