Institute of Science and Technology for Brain-Inspired Intelligence; MOE Key Laboratory of Computational Neuroscience and Brain-Inspired Intelligence; MOE Frontiers Center for Brain Science, Fudan University, Shanghai, 200433, China.
Tencent AI Lab, Shenzhen, China.
Nat Commun. 2022 Nov 28;13(1):7330. doi: 10.1038/s41467-022-34867-5.
The rapidly developing spatial omics generated datasets with diverse scales and modalities. However, most existing methods focus on modeling dynamics of single cells while ignore microenvironments (MEs). Here we present SOTIP (Spatial Omics mulTIPle-task analysis), a versatile method incorporating MEs and their interrelationships into a unified graph. Based on this graph, spatial heterogeneity quantification, spatial domain identification, differential microenvironment analysis, and other downstream tasks can be performed. We validate each module's accuracy, robustness, scalability and interpretability on various spatial omics datasets. In two independent mouse cerebral cortex spatial transcriptomics datasets, we reveal a gradient spatial heterogeneity pattern strongly correlated with the cortical depth. In human triple-negative breast cancer spatial proteomics datasets, we identify molecular polarizations and MEs associated with different patient survivals. Overall, by modeling biologically explainable MEs, SOTIP outperforms state-of-art methods and provides some perspectives for spatial omics data exploration and interpretation.
快速发展的空间组学产生了具有多种规模和模态的数据集。然而，大多数现有方法侧重于对单个细胞的动力学进行建模，而忽略了微环境（MEs）。在这里，我们提出了 SOTIP（空间组学多任务分析），这是一种将 MEs 及其相互关系纳入统一图的通用方法。基于这个图，可以进行空间异质性量化、空间域识别、差异微环境分析等下游任务。我们在各种空间组学数据集上验证了每个模块的准确性、鲁棒性、可扩展性和可解释性。在两个独立的小鼠大脑皮层空间转录组学数据集上，我们揭示了与皮层深度强烈相关的梯度空间异质性模式。在人类三阴性乳腺癌空间蛋白质组学数据集中，我们鉴定了与不同患者存活率相关的分子极化和 MEs。总的来说，通过对生物学上可解释的 MEs 进行建模，SOTIP 优于最先进的方法，并为空间组学数据的探索和解释提供了一些视角。
