湖南工业大学学报2025，Vol.39Issue(5)：52-57,6.DOI:10.3969/j.issn.1673-9833.2025.05.008
基于多尺度图对比学习的空间转录组聚类方法
A Spatial Transcriptome Clustering Method Based on Multi-Scale Map Contrast Learning
摘要
Abstract
In view of the flaw of discontinuous or intersecting spatial domains identified by graph neural networks in the clustering process of spatial transcriptome data,a spatial transcriptome clustering method mcmlST,which is based on multi-scale graph contrastive learning,has thus been proposed.Firstly,the spatial transcriptome data is preprocessed by using SCANPY and principal component analysis,followed by an enhancement of the ST data to form a new view.Next,based on graph autoencoders and auxiliary autoencoders,a dual encoding structure is designed to learn the embedded features of spatial transcriptome data.Finally,the k-means algorithm is used for an identification of spatial domains in spatial transcriptome data on the basis of embedded features.On three classic spatial transcriptome datasets(right dorso lateral prefrontal cortex,human breast cancer Block A Section 1 and STARmap),the proposed method calculates higher ARI and NMI compared with the three baseline methods conST,CCST,and DeepST,indicating a superior spatial transcriptome clustering performance.关键词
多尺度学习/多头注意力/对比图聚类/深度学习Key words
multi-scale learning/multi-head attention/contrast graph clustering/deep learning分类
计算机与自动化引用本文复制引用
阳龙,彭利红,周立前..基于多尺度图对比学习的空间转录组聚类方法[J].湖南工业大学学报,2025,39(5):52-57,6.基金项目
国家自然科学基金资助项目(62072172) （62072172）
