GCEA: Contrastive-Enhanced Autoencoders with Adaptive Completion for Partial Multi-omics Integration in Cancer Subtyping

Cancer subtype identification represents a crucial research domain in bioinformatics. The precise differentiation of patients into distinct cancer subtypes by leveraging information among multi-omics data represents a major challenge in the current research landscape. However, the presence of missing multi-omics data further complicates the precise identification of cancer subtypes. To conquer the limitation, this paper proposes a novel cancer subtype identification model, named GCEA, capable of addressing scenarios involving missing multi-omics data. Specifically, by leveraging adversarial generative strategies, the model adaptively generates missing omics data, thereby enabling it to flexibly handle various types of missing multi-omics data. Additionally, the model enhances its learning capability by employing contrastive learning to acquire informative and distinct latent representations of samples, facilitating precise identification of cancer subtypes in patients. Extensive experimental results on ten cancer multi-omics datasets demonstrate that the proposed model exhibits competitive performance compared with existing methods.