Metabolomic Profiling and Machine Learning Models for Tumor Classification in Patients with Recurrent IDH-Wild-Type...

Area of Science:

• Biochemistry
• Oncology
• Computational Biology

Background:

• Glioblastoma recurrence is a significant challenge in patient treatment.
• Identifying reliable biomarkers for tumor progression is crucial for managing recurrent disease.

Purpose of the Study:

• To identify the metabolomic signature associated with recurrent glioblastoma.
• To develop a predictive model for tumor phase using metabolomic data and machine learning.

Main Methods:

• Prospective blood sample analysis from six IDH-wildtype glioblastoma patients undergoing surgery at diagnosis and relapse.
• Untargeted gas chromatography-time-of-flight mass spectrometry for metabolite abundance measurement.
• Application of machine learning algorithms (gradient boosting, random forest) for predictive modeling.

Main Results:

• Significant changes in specific metabolites (e.g., decreased sorbitol, increased urea) were observed between pre-operative and post-relapse samples.
• Post-radiation analysis revealed decreased erythritol and 6-deoxyglucitol, and increased 2,4-difluorotoluene and 9-myristoleate.
• A gradient-boosting machine learning model demonstrated high performance in predicting tumor conditions post-relapse surgery.

Conclusions:

• A machine learning predictor for tumor phase was developed based on plasma metabolomic profiles.
• Metabolomics combined with machine learning shows potential for stratifying glioblastoma progression risk.