Abstract: Liver cancer is a common malignant tumor that threatens human health. To systematically acquire the knowledge of liver cancer by integrating genetic data using deep learning methods, we use a multi-omics disease analysis approach to explore the interrelationships between the groups and to obtain more accurate clinical decisions. However, due to the high dimensional sparsity of multi-omics data, there are a large number of redundant features and fewer available clinical label samples. Stacked denoising autoencoder (SDAE) is an efficient model that can obtain effective features from massive data. Therefore, based on the SDAE model, a hierarchical stacking denoising encoder was proposed to learn and integrate the characteristics of RNA expression, miRNA expression and DNA methylation data of liver cancer. The results show that the Hi-SDAE method improves the accuracy of the classification of liver cancer subtypes, and provides a more valuable reference for the targeted treatment of liver cancer.