基于多尺度邻域特征融合的空间转录组空间域识别研究  

Spatial domain identification of spatial transcriptomics based on multi-scale neighborhood feature fusion

作　　者：冯爱静 王文淼 高逸鸣 叶兰[2] 米加 赵小刚[4] 刘治[1] FENG Aijing;WANG Wenmiao;GAO Yiming;YE Lan;MI Jia;ZHAO Xiaogang;LIU Zhi(School of Information Science and Engineering,Shandong University,Qingdao 266237,China;The Cancer Center,the Second Hospital of Shandong University,Jinan 250012,China;Department of Special Examination,Shandong Provincial Third Hospital,Jinan 250002;Department of Thoracic Surgery,the Second Hospital of Shandong University,Jinan 250002)

机构地区：[1]山东大学信息科学与工程学院,青岛266237 [2]山东大学第二医院肿瘤防治中心,济南250012 [3]山东省立第三医院特检科,济南250002 [4]山东大学第二医院胸外科,济南250002

出　　处：《生物医学工程研究》2024年第6期468-474,共7页Journal Of Biomedical Engineering Research

基　　金：国家自然科学基金项目(11927808);山东省重点研发计划(重大科技创新项目)(2022CXGC010504);中央高校基本科研业务费专项资金(2022JC015);山东省自然科学基金资助项目(ZR2021LZH009);济南市“新高校20条”资助项目(2021GXRC024);青岛市科技惠民示范项目(24-1-8-cspz-20-nsh)。

摘　　要：为缓解空间转录组学数据因高离散性、稀疏性和分辨率受限导致空间域识别精度低的问题,基于多尺度邻域特征融合策略,本研究提出了一种空间域识别方法stMSAA。本研究在人类背外侧前额叶皮层数据集和人类乳腺癌数据集上进行实验,调整兰德系数、归一化互信息和聚类评价指数分别达到0.452、0.603和0.560。结果表明,stMSAA可有效整合多尺度的空间邻域信息,更准确地识别层次结构和边缘信息,提高空间域识别的精度和鲁棒性。To alleviate the problem of low spatial domain recognition accuracy due to high discreteness,sparsity and limited resolution of spatial transcriptomic data,based on multi-scale neighborhood feature fusion strategy,we proposed a spatial domain recognition method stMSAA.The experiment was conducted on the human dorsolateral prefrontal cortex dataset and the human breast cancer dataset.The adjusted rand index,normalized mutual information,and fowlkes-mallows index reached 0.452,0.603 and 0.560,respectively.The results show that stMSAA can effectively integrate different scale spatial neighborhood information,accurately identify hierarchical structures,edge information and improve the accuracy and robustness of spatial domain identification.

关 键 词：空间转录组 空间域识别 stAA 多尺度邻域 特征融合 

分 类 号：R318[医药卫生—生物医学工程] TP391[医药卫生—基础医学]