Artificial intelligence-based diagnostic model for identifying PTPRZ1-MET fusion in a clinically defined secondary glioblastoma cohort.
Summary
This study analyzed 159 patients with secondary glioblastoma (sGBM) and confirmed that the PTPRZ1-MET fusion is a recurrent oncogenic driver associated with significantly shorter overall and progression-free survival. Researchers identified 359 upregulated genes in fusion-positive tumors, enriched in cell-cycle regulation pathways. An artificial intelligence-based diagnostic model, utilizing a three-gene panel (MET, PCDHGA3, FAM3C), was developed and validated to identify this fusion. This robust molecular signature could facilitate the identification and stratification of sGBM patients positive for the PTPRZ1-MET fusion.
Analysis
CLINIQUE: This research highlights the PTPRZ1-MET fusion as an unfavorable prognostic biomarker in secondary glioblastoma patients, associated with significantly shorter overall and progression-free survival. The development of a diagnostic model based on a three-gene signature (MET, PCDHGA3, FAM3C) offers a promising path for more precise patient stratification. This could justify clinical trials specifically targeting patients with this fusion, potentially with MET inhibitors, and could influence clinical practice within 3-5 years by enabling better therapeutic selection. BIOMOL: The study discovered a three-gene molecular signature (MET, PCDHGA3, FAM3C) that acts as a surrogate biomarker for the PTPRZ1-MET fusion. This discovery was made via transcriptomic profiling and validated at the protein level using multiplex and conventional immunohistochemistry in FFPE samples, demonstrating good analytical validity for pathology detection. The ability to detect this signature in FFPE samples is crucial for clinical application, paving the way for a routine diagnostic test to identify patients eligible for targeted therapies. BIOINFO: An XGBoost classification model was employed to identify the most informative biomarkers, demonstrating a robust artificial intelligence approach for gene signature discovery. The model was rigorously validated through cross-validation, nested evaluation, feature-pool sensitivity analyses, and class-weighted modeling, ensuring its stable discriminative performance. If clinically deployed, this model could automate and enhance the accuracy of PTPRZ1-MET fusion diagnosis, thereby optimizing the workflow in molecular pathology laboratories.