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

PMCID: PMC12176387

PMID: 40530955

DOI: 10.7554/eLife.99849

Journal: eLife

Publication Date: 2025-6-18

Authors: Li K, Liang J, Li N, Fang J, Zhou X, et al.

Key Points

• NPC-RSS provides a robust, data-driven approach to predicting radiotherapy response in nasopharyngeal carcinoma
• Model achieved high predictive accuracy with AUC of 0.932 across combined datasets
• Identifies potential molecular targets for improving radiotherapy effectiveness and patient stratification

Summary

This study addresses a critical challenge in nasopharyngeal carcinoma (NPC) treatment by developing a novel predictive model, the Nasopharyngeal Carcinoma Radiotherapy Sensitivity Score (NPC-RSS), to identify patients' potential response to radiotherapy. Using an advanced machine learning approach combining glmBoost and NaiveBayes algorithms, researchers analyzed transcriptomic data to construct a predictive signature based on 18 key genes, demonstrating exceptional diagnostic performance with an area under the ROC curve of 0.996 in the training set and 0.823 in the external validation set.

The research provides comprehensive insights into radiotherapy sensitivity by revealing the molecular mechanisms underlying treatment response. The NPC-RSS not only predicts radiotherapy outcomes but also elucidates critical biological pathways, including immune-related signaling networks like Wnt/β-catenin, JAK-STAT, and NF-κB. Key genes such as SMARCA2, DMC1, and CD9 were identified as significant contributors to radiotherapy sensitivity, offering potential targets for personalized treatment strategies in NPC management.