A network embedding based method for partial multi-omics integration in cancer subtyping
Xu Han, Lin Gao, Mingfeng Huang, Ran Duan
Abstract
Integrative analysis of multiple omics offers the opportunity to uncover coordinated cellular processes acting across different omics layers. The ever-increasing of multi-omics data provides us a comprehensive insight into cancer subtyping. Many multi-omics integrative methods have been developed, but few of them can deal with partial datasets in which some samples only have data for a subset of the omics. In this study, we propose a partial multi-omics integrative method, MSNE (Multiple Similarity Network Embedding), for cancer subtyping. MSNE integrates the multi-omics information by embedding the neighbor relations of samples defined by the random walk on multiple similarity networks. We compared MSNE with five existing multi-omics integrative methods on twelve datasets in both full and partial scenarios. MSNE achieved the best result on pan-cancer and image datasets. Furthermore, on ten cancer subtyping datasets, MSNE got the most enriched clinical parameters and comparable log-rank test P-values in survival analysis. In conclusion, MSNE is an effective and efficient integrative method for multi-omics data and, especially, has a strong power on partial datasets.