Classification of High-Grade Serous Ovarian Cancer by Multi-Modal Deep Neural Networks

A molecular subtype classification model MMDNN-HGSOC is proposed, which integrates miRNA expression, DNA methylation, and copy number variation (CNV) with mRNA expression data to construct a multi-omics feature space. Based on LASSO (Least Absolute Shrinkage and Selection Operator) regression algorithm, a superposed LASSO (S-LASSO) regression algorithm is proposed to fully obtain gene subsets associated with HGSOC subtypes in each omics data. A late integration strategy for multi-omics data is introduced, and multi-modal deep neural networks are used to learn advanced feature representations of different omics data. The experimental results show that MMDNN-HGSOC performs well in the classification of HGSOC molecular subtypes. In addition, GO (Gene Ontology) and KEGG (Kyoto Encycloped Genomes) enrichment analyses are conducted on important genes discovered during feature selection, providing a strong support for the molecular subtype identification and pathogenesis research of HGSOC.