A Novel Approach for Subtype Identification via Multi-omics Data Using Adversarial Autoencoder

Cancer is a complex and heterogeneous disease, and effective diagnosis and treatment require accurate description of tumor subtypes. Traditional cancer identification methods based on clinical and histopathological criteria have limitations in identifying crucial molecular subtypes. In recent years, high-throughput genomics technologies have changed cancer research and made it possible to characterize tumor molecular features on a large scale. The integration of multiple genomic data types is expected to provide a more comprehensive understanding of the potential molecular mechanisms of cancer and facilitate the discovery of new diagnostic and therapeutic targets, but new computational methods need to be developed. In this study, we propose MultiDAAE (Multi-omics Double Adversarial Autoencoder) by combining autoencoders with two discriminators, which forms two generative adversarial networks and is able to more effectively extract features and reduce dimensionality of multi-omics data. Compared to state-of-the-art methods, our approach exhibits excellent clustering performance in multiple cancer datasets. In summary, MultiDAAE can help reveal potential molecular mechanisms and provide insights for the development of personalized cancer therapies.