Multi-omics clustering based on dual contrastive learning for cancer subtype identification

Multi-omics clustering aims to supplement a single omics with data information from multiple omics, assigning samples into their respective clusters without supervision. The existing multi-omics clustering methods tend to use the similarity measure function to construct the similarity network between samples, and then fuse the various omics networks together for clustering by some fusion methods. These methods relies heavily on the original features of the data and the similarity measure function. This paper proposes a novel multi-omics clustering method. The proposed model first uses neural networks to extract the feature embeddings of omics, and then aligns the feature embeddings of different omics in subspaces through contrastive learning. Finally, the feature embeddings are mapped to clustered soft labels, and the soft labels are aligned again using contrastive learning. Experiments show that our method outperforms the baseline methods.