Development and validation of a robust cuproptosis related signature for primary glioma via machine learning aided by loop training and validation.

BACKGROUND: Glioma is the most common malignant tumors in central nervous system with high mortality. Accurately predicting prognosis for patients with glioma still remains a challenge. Accumulated studies have found that cuproptosis-related genes emerged as potential biomarkers for cancer prognosis. However, their prognostic roles in primary glioma are unclear. This study aimed to develop a promising prognostic signature for primary glioma using cuproptosis-related genes. METHODS: A total of 1248 patients with primary glioma were obtained from The Cancer Genome Atlas (TCGA) and Chinese Glioma Genome Atlas (CGGA) databases. 101 machine learning algorithm combinations together with a loop training and validation procedure were performed to identify the optimal model termed cuproptosis-related prognostic signature (CRPS). The predictive accuracy of CRPS was evaluated through Kaplan-Meier survival curves and receiver-operator characteristic (ROC) analyses. Furthermore, we compared the performance of CRPS with common clinical features and 72 published prognostic signatures. RESULTS: CRPS exhibited robust predictive capability in overall survival (OS) and could serve as an independent prognostic biomarker in different cohorts including TCGA-GBMLGG (HR: 1.987, 95%CI: 1.239-3.189, p < 0.001), CGGA693 (HR: 2.374, 95%CI: 1.505-3.745, p < 0.001) and CGGA325 (HR: 2.248, 95%CI: 1.334-3.787, p = 0.002). Simultaneously, CRPS outperformed 72 published signatures and traditional clinical features. Additionally, a nomogram by the combination of CRPS and tumor grade contributes to more precise prognosis prediction. CONCLUSIONS: Our study highlights CRPS as a promising tool in prognosis evaluation, survival risk stratification and personalized clinical management for primary glioma.