Dong Mingze, Su David G, Kluger Harriet, Fan Rong, Kluger Yuval
Interdepartmental Program in Computational Biology & Bioinformatics, Yale University, New Haven, CT, USA.
Department of Pathology, Yale School of Medicine, New Haven, CT, USA.
Nat Commun. 2025 Mar 27;16(1):2990. doi: 10.1038/s41467-025-58089-7.
Spatial omics technologies enable analysis of gene expression and interaction dynamics in relation to tissue structure and function. However, existing computational methods may not properly distinguish cellular intrinsic variability and intercellular interactions, and may thus fail to reliably capture spatial regulations. Here, we present Spatial Interaction Modeling using Variational Inference (SIMVI), an annotation-free deep learning framework that disentangles cell intrinsic and spatial-induced latent variables in spatial omics data with rigorous theoretical support. By this disentanglement, SIMVI enables estimation of spatial effects at a single-cell resolution, and empowers various downstream analyses. We demonstrate the superior performance of SIMVI across datasets from diverse platforms and tissues. SIMVI illuminates the cyclical spatial dynamics of germinal center B cells in human tonsil. Applying SIMVI to multiome melanoma data reveals potential tumor epigenetic reprogramming states. On our newly-collected cohort-level CosMx melanoma data, SIMVI uncovers space-and-outcome-dependent macrophage states and cellular communication machinery in tumor microenvironments.
空间组学技术能够分析与组织结构和功能相关的基因表达及相互作用动态。然而，现有的计算方法可能无法恰当地区分细胞内在变异性和细胞间相互作用，因此可能无法可靠地捕捉空间调控。在此，我们提出了使用变分推理的空间相互作用建模（SIMVI），这是一个无需注释的深度学习框架，在严格的理论支持下，它能够解开空间组学数据中的细胞内在和空间诱导的潜在变量。通过这种解缠，SIMVI能够在单细胞分辨率下估计空间效应，并支持各种下游分析。我们展示了SIMVI在来自不同平台和组织的数据集上的卓越性能。SIMVI揭示了人类扁桃体中生发中心B细胞的周期性空间动态。将SIMVI应用于多组学黑色素瘤数据，揭示了潜在的肿瘤表观遗传重编程状态。在我们新收集的队列水平的CosMx黑色素瘤数据上，SIMVI揭示了肿瘤微环境中依赖空间和结果的巨噬细胞状态及细胞通讯机制。
