Weighted gene co-expression network analysis combined with machine learning to identify prognostic biomarkers for cervical squamous cell carcinoma.

Abstract

e17000 Background: Cervical cancer represents the fourth most frequently diagnosed malignancy affecting women all over the world. However, effective prognostic biomarkers are still limited for accurate identifying high-risk patients. Here, we provide a co-expression network and machine learning-based signature to predict the survival of cervical cancer. Methods: Utilizing expression profiles of The Cancer Genome Atlas datasets, we identified differentially expressed genes (DEGs) and the most significantly module by differential expression analysis and Weighted Gene Co-expression Network Analysis, respectively. The candidate genes was obtained by combining the both results. Then the prognostic classifier was constructed by LASSO COX regression analysis and validated in testing set. Finally, survival receiver operating characteristic and Cox proportional hazards analysis was used to assess the performance of prognostic prediction. Results: We identified 190 differentially expressed genes (DEGs) between cervical squamous cell cancer(CSCC) and normal samples in purple module. Next we built a 8-mRNA-based signature, and determined a optimal cutoff value with sensitivity of 0.889 and specificity of 0.785. Patients were classified into high-risk and low-risk group with significantly different overall survival(training set: p < 0.0001; testing set: p = 0.039). Furthermore, the prognostic classifier was an independent and powerful prognostic biomarker for OS (HR = 7.05, 95% CI: 2.52-19.71, p < 0.001). Conclusions: The prognostic classifier is a promising predictor of CSCC patients, the novel co-expression network and machine learning-based strategy described in the study may have a broad application in precision medicine.