Suppr超能文献

stClinic通过在动态图中整合空间多层多组学数据来剖析临床相关生态位。

stClinic dissects clinically relevant niches by integrating spatial multi-slice multi-omics data in dynamic graphs.

作者信息

Zuo Chunman, Xia Junjie, Xu Yupeng, Xu Ying, Gao Pingting, Zhang Jing, Wang Yan, Chen Luonan

机构信息

School of Life Sciences, Sun Yat-sen University, Guangzhou, China.

Institute of Artificial Intelligence, Donghua University, Shanghai, China.

出版信息

Nat Commun. 2025 Jun 16;16(1):5317. doi: 10.1038/s41467-025-60575-x.

Abstract

Spatial multi-slice multi-omics (SMSMO) integration has transformed our understanding of cellular niches, particularly in tumors. However, challenges like data scale and diversity, disease heterogeneity, and limited sample population size, impede the derivation of clinical insights. Here, we propose stClinic, a dynamic graph model that integrates SMSMO and phenotype data to uncover clinically relevant niches. stClinic aggregates information from evolving neighboring nodes with similar-profiles across slices, aided by a Mixture-of-Gaussians prior on latent features. Furthermore, stClinic directly links niches to clinical manifestations by characterizing each slice with attention-based geometric statistical measures, relative to the population. In cancer studies, stClinic uses survival time to assess niche malignancy, identifying aggressive niches enriched with tumor-associated macrophages, alongside favorable prognostic niches abundant in B and plasma cells. Additionally, stClinic identifies a niche abundant in SPP1+ MTRNR2L12+ myeloid cells and cancer-associated fibroblasts driving colorectal cancer cell adaptation and invasion in healthy liver tissue. These findings are supported by independent functional and clinical data. Notably, stClinic excels in label annotation through zero-shot learning and facilitates multi-omics integration by relying on other tools for latent feature initialization.

摘要

空间多切片多组学(SMSMO)整合改变了我们对细胞微环境的理解,尤其是在肿瘤方面。然而,诸如数据规模和多样性、疾病异质性以及样本量有限等挑战,阻碍了临床见解的得出。在此,我们提出了stClinic,这是一种动态图模型,它整合了SMSMO和表型数据以揭示临床相关的微环境。stClinic借助潜在特征上的高斯混合先验,聚合来自跨切片具有相似特征的不断演化的相邻节点的信息。此外,stClinic通过相对于总体使用基于注意力的几何统计量来表征每个切片,将微环境直接与临床表现联系起来。在癌症研究中,stClinic使用生存时间来评估微环境的恶性程度,识别富含肿瘤相关巨噬细胞的侵袭性微环境,以及富含B细胞和浆细胞的有利预后微环境。此外,stClinic识别出一个在SPP1 + MTRNR2L12 + 髓样细胞和癌症相关成纤维细胞中丰富的微环境,该微环境驱动结直肠癌细胞在健康肝组织中的适应和侵袭。这些发现得到了独立的功能和临床数据的支持。值得注意的是,stClinic在通过零样本学习进行标签注释方面表现出色,并通过依赖其他工具进行潜在特征初始化来促进多组学整合。

https://cdn.ncbi.nlm.nih.gov/pmc/blobs/a1de/12170857/4a3322d8c790/41467_2025_60575_Fig1_HTML.jpg

文献AI研究员

20分钟写一篇综述,助力文献阅读效率提升50倍。

立即体验

用中文搜PubMed

大模型驱动的PubMed中文搜索引擎

马上搜索

文档翻译

学术文献翻译模型,支持多种主流文档格式。

立即体验