Subtyping and grading of lower-grade gliomas using integrated feature selection and support vector machine.

Area of Science:

• Oncology
• Bioinformatics
• Computational Biology

Background:

• Accurate classification of lower-grade gliomas (LGGs) is vital for effective treatment.
• Histopathological classification of LGG subtypes faces significant interobserver variability, impacting patient care.
• Transcriptome data offers a molecular basis for improved diagnostic accuracy.

Purpose of the Study:

• To develop and validate a machine learning framework for classifying LGG subtypes and grades.
• To investigate the impact of cancer grade on classification accuracy.
• To identify novel predictive biomarkers for LGG.

Main Methods:

• An integrated feature selection method combining correlation and Support Vector Machine (SVM) recursive feature elimination was developed.
• An SVM classifier was implemented and compared against other machine learning frameworks.
• Differential co-expression analysis, gene set enrichment, and survival analysis were used for biological interpretation and biomarker discovery.

Main Results:

• The SVM classifier achieved superior accuracy in LGG classification compared to other methods.
• Subtype classification accuracy exceeded 90% within specific grades, while mixed-grade classification accuracy was around 80%.
• Higher heterogeneity in mixed-grade cancers was linked to reduced prediction accuracy. The six-class model achieved an average accuracy of 91%.

Conclusions:

• Accurate classification of both cancer grade and subtype is essential for improving diagnostic precision in LGGs.
• The developed machine learning framework demonstrates high accuracy and biological interpretability.
• The findings suggest that this framework can potentially assist clinicians in diagnosing and treating LGGs.