Multi-Omics-Based Subtyping of Ovarian Cancer Using Deep Learning and SVM Classification

Ovarian Cancer is considered to have a poor prognosis since it has a very heterogeneous nature. Hence, it is crucial to recognize its molecular subtypes to enhance its prognosis and therapy. Comparing with using a single omics data, multi-omics data analysis can provide a broader range of information and hence ensure an accurate identification of subtypes. This research aims to identify subtypes of ovarian cancer by using multi-omics data from 296 samples and develop a support vector machine (SVM) classifier to predict the subtypes for new patients. The subtypes are found using deep learning methods like Autoencoder and Consensus K-means Clustering. Two clusters were resulted differing significantly in the abundance of activated Mast Cells with p-value = 0.0430. Furthermore, enrichment analysis specified that genes such as CDC42EP5, EHD2, CLTB, NPDC1, and BASP1 were upregulated in subtype 2 than subtype 1, whereas DDOST, SUMF2, IGFBP2, CALR, and APMAP were downregulated. Pathways, including RHO GTPases, activate PAKs, and Signaling by WNT were upregulated, whereas Neutrophil degranulation and Mitotic Anaphase were downregulated. The SVM classifier achieved an accuracy of 0.94 ± 0.03, demonstrating its high performance.