Prediction of Patients With High-Risk Osteosarcoma on the Basis of XGBoost Algorithm Using Transcriptome and Methylation Data From SGH-OS Cohort

• DOI: 10.1200/po-24-00732
• Publication Date: 2025-03-28T20:01:10Z
• Authors: Weisong Zhao; Huanliang Meng; Zhenwu Dai; Lulu Zhang; Zhiwei Cheng; Yunjie Song; Wenyuan Xu; Zhuoying Wang; Kai Tian; Yafei Jiang; Wei Sun; Zhengdong Cai; Gangyang Wang; Yingqi Hua

PURPOSE Osteosarcoma (OS) is the most prevalent primary malignant bone sarcoma, characterized by its high rates of metastasis and mortality. In our previous multiomics analysis of the Shanghai General Hospital OS (SGH-OS) cohort, we identified four distinct OS subtypes, each with unique molecular characteristics and clinical outcomes. Of particular importance was the identification of the MYC-driven subtype, which exhibited the poorest prognosis and was referred to as high-risk OS. A diagnostic tool is needed for clinicians to identify high-risk OS in advance. The purpose of this study is to develop a classifier capable of accurately predicting the high-risk OS subtype using transcriptome and methylation data.

METHODS In this study, using eXtreme Gradient Boosting (XGBoost) with Bayesian optimization, we developed a classification model by integrating transcriptome and methylation data from our internal SGH-OS cohort. We further validated the model's predictive performance with the external TARGET-OS cohort.

RESULTS Using the XGBoost algorithm, we developed a classifier incorporating nine genes (ARHGAP9, CADM1, CPE, DUSP3, FGFR1, GALNT3, IGF2BP3, KIF26A, ZFP3). In our internal cohort, the classifier exhibited excellent predictive performance, with an area under the receiver operating characteristics curve (AUC) of 0.999 and an overall accuracy of 0.989. Furthermore, the classifier successfully stratified two groups with distinct survival outcomes in the external TARGET-OS cohort. Notably, our analysis revealed a positive correlation between IGF2BP3 and MYC signaling pathways, highlighting IGF2BP3 as a potential therapeutic target in high-risk OS.

CONCLUSION Our classifier demonstrated excellent predictive performance in identifying patients with high-risk OS, offering the potential to enhance treatment decision making and optimize patient management strategies.