Identification of novel biomarkers for epithelial ovarian cancer through machine learning and explainable artificial intelligence using in silico and in vitro analysis

Epithelial ovarian cancer (EOC) is a lethal gynecological malignancy. Ongoing research aimed to identify novel biomarkers and develop combined algorithms to improve diagnosis and prognosis prediction for EOC. RNA-seq related to EOC were obtained from the Recount 3 databases data. We categorized early, late stages versus control, all stages (I-IV) versus control, and early versus late stages for machine learning, deep learning model and explainable AI (XAI). The hub genes were selected by considering SHapley Additive exPlanations (SHAP) analysis. The expression of hub genes was validated in fresh tissue (34 EOC, and 33 benign neoplastic ovarian) and peripheral blood mononuclear cell (PBMC) from 39 EOC, and 33 benign neoplastic ovarian patients using real time-PCR. For late stage and also all stages versus control, three hub genes were selected including SGO1, VTA1, RBM5-AS1 were mainly involved in cell cycle, estrogen signaling and vesicle trafficking. SGO1, and VTA1 were increased significantly in both tissue and PBMC of EOC patients (p < 0.05). RBM5-AS1 was significantly decreased in EOC tissue as well as PBMC compared to benign neoplastic ovarian patients (p < 0.05). VTA1 and RBM5-AS1 show promise as both diagnostic and prognostic markers, while SGO1 may serve primarily as a diagnostic biomarker for EOC.