MOFNet: A Deep Learning Framework of Integrating Multi-omics Data for Breast Cancer Diagnosis

With the advancement of technology, annotated multi-omics datasets are becoming increasingly abundant. In this paper, we propose a novel deep learning framework, called multi-omics data fusion network (MOFNet), to integrate multi-omics data for disease diagnosis. MOFNet is a multi-task learning framework that combines multiple deep learning models to learn the complex relationships between multi-omics data and disease label. MOFNet focuses on improving disease classification performance with fewer features extracted from interrelated multi-omics data. We demonstrate that MOFNet outperforms other state-of-the-art supervised multi-omics data integration methods in breast cancer sample classification tasks using mRNA expression, DNA methylation, and microRNA expression profiles. The selected features can be regarded as integrative biomarkers of breast cancer diagnosis and stratification.