乳腺癌空间转录组数据集上基于深度学习的EnST算法研究
EnST algorithm based on deep learning on breast cancer spatial transcriptome dataset
赵雅楠 1尹娜 1司志好 1尚文婧 1冯振兴1
作者信息
- 1. 内蒙古工业大学 理学院,呼和浩特 010051
- 折叠
摘要
在保留空间位置和组织学图像的基础上,空间转录组学使用基因表达谱数据对组织结构和生物发育提出新的见解.准确识别位点的空间域是空间转录组学各种下游分析的重要步骤,提出的EnST算法在运用复合缩放网络的基础上,添加了变分图自编码器,能够在空间转录组数据中提取有用信息.在人类乳腺癌空间转录组学数据集中,相比于其他算法,EnST算法可以更好地描绘乳腺癌精细的空间组织结构.此外,EnST学习到的表征在聚类、可视化、差异基因表达分析、GO功能分析等下游任务中也展现出强大的性能.
Abstract
Based on preserving spatial location and histological images,the spatial transcriptome can provide new insights into tissue structure and organism development with the data of gene expression profile.To precisely identify the spatial domain of loci is an important step in various downstream analyses of spatial transcriptome.The proposed EnST algorithm adds a variogram self-encoder based on the use of composite scaling networks,which can extract useful information in spatial transcriptome data.In the spatial transcriptome dataset of human breast cancer,the EnST algorithm can made a better description on the fine spatial organization structure of breast cancer in comparison with other seven algorithms.In addition,the representations learned by EnST showed a powerful performance on the downstream tasks of clustering,visualization,differential gene expression analysis and GO function analysis.
关键词
深度学习/变分图自编码器/乳腺癌/空间域/聚类Key words
deep learning/variational graph autoencoder/breast cancer/spatial domain/clustering引用本文复制引用
基金项目
内蒙古自治区自然科学基金(2019BS03025)
出版年
2024
NETL
NSTL
万方数据