Machine learning model predicts radiotherapy response in patients with nasopharyngeal carcinoma

Researchers in China have developed a powerful machine learning model that can help determine which patients with nasopharyngeal carcinoma (NPC) are likely to respond well to radiotherapy-a common treatment for this type of cancer. The study, conducted by scientists at Zhujiang Hospital and Nanfang Hospital of Southern Medical University, introduces a predictive tool known as the NPC-RSS (Nasopharyngeal Carcinoma Radiotherapy Sensitivity Score).

Using transcriptomic data and a rigorous machine learning framework that evaluated 113 algorithm combinations, the team identified an 18-gene signature capable of predicting a patient's radiosensitivity. The model showed impressive accuracy in both internal datasets and external validation sets.

Radiotherapy is the primary treatment for NPC, but up to 30% of patients relapse due to radiation resistance. Our model helps solve this problem by identifying patients who are most likely to benefit from radiotherapy, allowing for more tailored and effective treatment strategies."

Dr. Jian Zhang, lead author 

The model's core genes-such as SMARCA2, DMC1, and CD9-were found to influence tumor immune infiltration and key signaling pathways like Wnt/β-catenin and JAK-STAT. Notably, the radiosensitive group showed higher levels of immune cell activity, suggesting an intimate connection between radiation response and immune dynamics.

The predictive power of the NPC-RSS was confirmed using cell lines and single-cell sequencing, showing that radiosensitive tumors have richer immune environments compared to resistant ones. According to co-author Dr. Hui Meng, "Our findings suggest that integrating gene scores with immune profiles could be a game-changer in NPC care."

The team believes the model could become a clinical tool for guiding treatment decisions, minimizing unnecessary radiation exposure, and optimizing therapeutic outcomes. They are now working to expand their sample size and collaborate with international partners to further validate and refine the model.