Drug Repurposing in Glioblastoma Using a Machine Learning-Based Hybrid Feature Selection Approach.

Area of Science:

• Oncology
• Bioinformatics
• Machine Learning

Background:

• Glioblastoma (GBM) is an aggressive brain cancer with poor prognosis and limited treatment options.
• Effective drug sensitivity prediction is crucial for developing targeted therapies.

Purpose of the Study:

• To apply a hybrid feature selection method for categorizing drug sensitivity in GBM.
• To identify discriminative drug compound features for improved classification performance and model interpretability.

Main Methods:

• A hybrid feature selection approach combining two popular methods with a rank-based weighting scheme was employed.
• Machine learning (ML) was utilized to analyze Genomics of Drug Sensitivity in Cancer (GDSC) datasets.
• The method focused on reducing dimensionality in high-dimensional drug sensitivity data.

Main Results:

• The ML-driven feature selection achieved over 95% accuracy in classifying drug sensitivity features.
• The approach successfully identified a small set of highly discriminative features (≤11 per metric).
• Improved model stability and performance were observed for drug compound-based predictions.

Conclusions:

• The developed feature selection approach enhances the precision and interpretability of GBM drug sensitivity prediction models.
• This method offers a promising direction for clinically actionable advancements in glioblastoma research.
• The findings pave the way for more targeted and effective GBM therapies.