A multi-gene predictive model for the radiation sensitivity of nasopharyngeal carcinoma based on machine learning.

Abstract

Radiotherapy resistance in nasopharyngeal carcinoma (NPC) is a major cause of recurrence and metastasis. Identifying radiotherapy-related biomarkers is crucial for improving patient survival outcomes. This study developed the nasopharyngeal carcinoma radiotherapy sensitivity score (NPC-RSS) to predict radiotherapy response. By evaluating 113 machine learning algorithm combinations, the glmBoost+NaiveBayes model was selected to construct the NPC-RSS based on 18 key genes, which demonstrated good predictive performance in both public and in-house datasets. The study found that NPC-RSS is closely associated with immune features, including chemokine factors and their receptor families and the major histocompatibility complex (MHC). Gene functional analysis revealed that NPC-RSS influences key signaling pathways such as Wnt/β-catenin, JAK-STAT, NF-κB, and T cell receptors. Cell line validation confirmed that SMARCA2 and CD9 gene expression is consistent with NPC-RSS. Single-cell analysis revealed that the radiotherapy-sensitive group exhibited richer immune infiltration and activation states. NPC-RSS can serve as a predictive tool for radiotherapy sensitivity in NPC, offering new insights for precise screening of patients who may benefit from radiotherapy.