A machine learning-based basement membrane gene signature model for predicting ovarian cancer survival.

Background: Ovarian cancer is a highly invasive malignancy that lacks early symptoms. The basement membrane, which separates epithelial and stromal tissues, is highly implicated in tumor development and invasion. Aberrant expression of basement membrane genes is associated with tumor cell infiltration, invasion, and poor prognosis. This study developed a machine learning-based basement membrane gene signature (BMGS) model for predicting the prognosis of patients with ovarian cancer. Methods: Transcriptomic data, clinical data, and the status of 222 basement membrane genes were retrieved from The Cancer Genome Atlas (TCGA), Gene Expression Omnibus (GEO), Genotype-Tissue Expression (GTEx) Project, and basement membraneBASE. After filtering zero-expression genes, we identified differentially expressed genes (P<0.05, and |log2 fold change| >0.585). We selected tumor-related genes, and 7 machine learning algorithms [including extreme gradient boosting (XGBoost)] with 10-fold cross-validation were used to construct the BMGS model, which was validated via Kaplan-Meier curves, receiver operating characteristic (ROC) analysis, and Cox regression. Results: In the multivariate Cox regression analyses, both the TCGA training set (P<0.001) and the GEO validation set (P=0.005) consistently demonstrated that the model was an independent risk factor for ovarian cancer prognosis. The BMGS-high group was associated with significantly higher aneuploidy scores (P<0.001) and higher frequency of Conclusions: This study confirms that the XGBoost-based BMGS with 39 core genes is an independent prognostic factor for ovarian cancer (TCGA: P<0.001; GEO: P=0.005). High BMGS risk correlates with significantly elevated aneuploidy (P<0.001) and