Background: High-grade serous ovarian cancer (HGSOC) is the second most lethal gynecologic malignancy, often diagnosed at a late stage due to the lack of reliable early detection strategies. Currently, there are no specific diagnostic or prognostic biomarkers for ovarian cancer (OC). Thus, there is a great need for novel validated biomarkers for OC diagnosis.
Methods: A two-step machine learning approach was employed to identify potential HGSOC biomarkers in The Cancer Genome Atlas (TCGA) and Genotype-Tissue Expression project (GTEx) ovarian cohorts. The selected genes were then validated in an external, clinically annotated tissue cohort of 65 samples from OC patients, treated in Lithuania, to assess biomarker performance in separating HGSOC from benign gynecologic conditions and predict overall survival.
Results: A ten-gene signature (EXO1, RAD50, PPT2, LUC7L2, PKP3, CDCA5, ZFPL1, VPS33B, GRB7, and TCEAL4) was selected for further analysis in the external cohort. Expression of each of the ten genes was highly indicative of HGSOC compared to benign gynecologic conditions (p ≤ 0.030) and separated these groups with GRB7 expression reaching the highest area under the ROC curve (AUC) of 0.986. RAD50, VPS33B, and ZFPL1 expression also correlated with stage in HGSOC cases (p < 0.042), while TCEAL4 expression was associated with tumor grade (p = 0.038). The 10-gene signature was also predictive of 5-year survival in the OC tissue cohort (AUC = 0.815).
Conclusions: The ten selected gene expression biomarkers could be useful for HGSOC diagnosis and prognosis; however, further investigations in their prediction of OC patients’ survival are still required.
Supplementary Information: The online version contains supplementary material available at 10.1186/s13048-026-01989-z.
Keywords: Diagnostic model; Gene expression; High-grade serous ovarian carcinoma; Ovarian cancer.