] were obtained, where each set consists of genes either up or down-regulated in each breast cancer subtype. For PAM50 genes, during the multi-omics data integration step, most of the genes were filtered out, and only 18 genes were left as the input for the classifier. As a result, 53 genes of the top 200 genes were the functional genes known for the breast cancer subtype .
The major contribution of our study resides within the multi-omics data integration strategy. To maintain the biological relationship between the multi omics features while integration, feature selection module was constructed to identify the informative breast cancer signature genes and the relation between the identified genes and other two omics features were built based on the promoter and the target relationship.