Identification of Prognostic Biomarkers in Gene Expression Profile of Neuroblastoma Via Machine Learning.

Abstract

Neuroblastoma (NB) is a common pediatric solid malignancy characterized by heterogeneous clinical outcomes. The identification of predictive and interpretable prognostic biomarkers is critical for advancing precision medicine in NB. We proposed an integrative network-based machine learning method for biomarker discovery, which employed a network smoothed t-statistic support vector machine to select prognostic related biomarkers, and then we performed network analysis on these biomarkers to find hub genes. Later, we conducted a comprehensive analysis to integrate bulk and single-cell RNA sequencing data to character the tumor microenvironment of prognostic state and correlated them to the discovered hub genes. This analysis identified 528 prognostic biomarkers associated with NB. Network-based analysis further refined this set to 11 hub prognostic biomarkers for NB: AURKA, BLM, BRCA1, BRCA2, CCNA2, CHEK1, E2F1, MAD2L1, PLK1, RAD51, and RFC3. Among these genes, high RFC3 expression was significantly associated with poor prognosis, highlighting its potential as a novel prognostic biomarker in NB. Additionally, our findings revealed that these biomarkers are correlated to chemotherapy drugs, such as vincristine and cyclophosphamide. Furthermore, drug sensitivity analyses identified several candidate drugs, such as dactinomycin, bortezomib, docetaxel, and sepantronium bromide, that may hold therapeutic potential for NB treatment. This study offers novel insights to underlying NB prognosis and therapeutic targets and provides a foundation for developing personalized treatment strategies to improve clinical outcomes.