Machine learning and experimental validation to construct a metastasis-related gene signature and ceRNA network for predicting osteosarcoma prognosis

Objective: Osteosarcoma (OS) is more common in adolescents and significantly harmful, and the survival rate is considerably low, especially in patients with metastatic OS. The identification of effective biomarkers and associated regulatory mechanisms, which predict OS occurrence and development as well as improve prognostic accuracy, will help develop more refined protocols for OS treatment. Methods: In this study, genes showing differential expression in metastatic and non-metastatic types of OS were identified, and the ones affecting OS prognosis were screened from among these. Following this, the functions and pathways associated with the genes were explored via enrichment analysis, and an effective predictive signature was constructed using Cox regression based on the machine learning algorithm, least absolute shrinkage and selection operator (LASSO). Next, a correlative competing endogenous RNA (ceRNA) regulatory axis was constructed after verification by bioinformatics analysis and luciferase reporter gene experiments conducted based on the prognostic signature. Results: Overall, 251 differentially expressed genes were identified and screened using bioinformatics and double luciferase reporter gene experiments. An effective prognostic signature was constructed based on 15 genes associated with OS metastasis, and upstream non-coding RNAs were identified to construct the "NBR2/miR-129-5p/FKBP11 " regulatory axis based on the ceRNA networks, which helped identify candidate biomarkers for the OS clinical diagnosis and treatment, drug research, and prognostic prediction, among other applications. The findings of this study provide a novel strategy for determining the mechanism underlying OS occurrence and development and the appropriate treatment.