Li Juntao, Xiang Shan, Wei Dongqing
School of Mathematics and Statistics, Henan Normal University, 46 Jianshe East Road, 453007 Xinxiang, China.
School of Life Sciences and Biotechnology, Shanghai Jiao Tong University, 800 Dongchuan Road, 200240 Shanghai, China.
Brief Bioinform. 2024 Nov 22;26(1). doi: 10.1093/bib/bbae707.
Identifying spatial domains is critical for understanding breast cancer tissue heterogeneity and providing insights into tumor progression. However, dropout events introduces computational challenges and the lack of transparency in methods such as graph neural networks limits their interpretability. This study aimed to decipher disease progression-related spatial domains in breast cancer spatial transcriptomics by developing the three graph regularized non-negative matrix factorization (TGR-NMF). A unitization strategy was proposed to mitigate the impact of dropout events on the computational process, enabling utilization of the complete gene expression count data. By integrating one gene expression neighbor topology and two spatial position neighbor topologies, TGR-NMF was developed for constructing an interpretable low-dimensional representation of spatial transcriptomic data. The progressive lesion area that can reveal the progression of breast cancer was uncovered through heterogeneity analysis. Moreover, several related pathogenic genes and signal pathways on this area were identified by using gene enrichment and cell communication analysis.
识别空间域对于理解乳腺癌组织异质性以及洞察肿瘤进展至关重要。然而，数据缺失事件带来了计算挑战，并且诸如图神经网络等方法缺乏透明度，限制了它们的可解释性。本研究旨在通过开发三图正则化非负矩阵分解（TGR-NMF）来破译乳腺癌空间转录组学中与疾病进展相关的空间域。提出了一种归一化策略来减轻数据缺失事件对计算过程的影响，从而能够利用完整的基因表达计数数据。通过整合一种基因表达邻域拓扑和两种空间位置邻域拓扑，开发了TGR-NMF用于构建空间转录组数据的可解释低维表示。通过异质性分析揭示了能够反映乳腺癌进展的渐进性病变区域。此外，利用基因富集和细胞通讯分析确定了该区域的几个相关致病基因和信号通路。
