基于归一化欧氏距离和谱分邻的融合网络对癌症亚型聚类

Clustering of Cancer Subtypes Based on Similarity Network Fusion of Normalized Euclidean Distance and Spectral Neighbor

摘要: 相似网络融合(SNF)是一种确定癌症亚型的有效聚类方法，可以将病人不同数据类型的相似网络融合成一个相似网络，融合后的相似网络包含病人的所有信息。本文使用归一化的欧氏距离和重新定义的病人的邻居，提出了基于归一化欧氏距离和谱聚类分邻的相似网络融合(NSSNF)，减少了相似网络中数据分析时产生的噪声，并增加了不同相似网络的数据间的互补性，最后，针对数据库TCGA中的五种癌症数据，利用NSSNF方法进行数据分析，评价指标DB和CH的结果表明NSSNF方法优于传统SNF方法、NSNF方法和CSNF方法。

Abstract: Similarity network fusion (SNF) is an effective clustering method to identify cancer subtypes. By us-ing SNF, patient similarity networks for each of their data types are integrated into a single similari-ty network, which contains all the information of patients. In this paper, a similarity network fusion based on normalized Euclidean distance and spectral clustering (NSSNF) is proposed by using nor-malized Euclidean distance and redefined neighbor of the patient to reduce the noise of data analy-sis in similarity networks and increase the complementarity between the data from different simi-larity networks. Finally, for the five cancer data types from the TCGA database, the data analysis was performed by the NSSNF method, and the results of the evaluation indexes DB and CH showed that the NSSNF method is superior to the SNF method, NSNF method and CSNF method.