A Hybrid Machine Learning Approach to Screen Optimal Predictors for the Classification of Primary Breast Tumors from...

Area of Science:

• Bioinformatics
• Genomics
• Computational Biology

Background:

• Microarray gene expression data presents challenges in dimensionality and sparsity for breast cancer (BC) analysis.
• Identifying optimal gene subsets for BC prediction is crucial for accurate diagnosis.

Purpose of the Study:

• To develop a novel hybrid Feature Selection (FS) framework for identifying optimal gene biomarkers for BC prediction.
• To evaluate the diagnostic capability of selected gene biomarkers using various supervised Machine Learning (ML) algorithms.

Main Methods:

• A hybrid FS framework combining minimum Redundancy-Maximum Relevance (mRMR), unpaired t-test, and meta-heuristics was employed.
• Three optimal gene biomarkers (MAPK 1, APOBEC3B, ENAH) were identified.
• Supervised ML algorithms including XGBoost, SVM, KNN, NN, NB, DT, and LR were used to assess predictive performance.

Main Results:

• The framework successfully identified MAPK 1, APOBEC3B, and ENAH as key gene biomarkers for BC.
• The eXtreme Gradient Boosting (XGBoost) model demonstrated superior performance with high accuracy (0.976 ± 0.027), F1-Score (0.974 ± 0.030), and AUC (0.961 ± 0.035).
• The developed classification system effectively distinguishes primary breast tumors from normal samples.

Conclusions:

• The proposed hybrid FS framework is effective in identifying optimal gene biomarkers for breast cancer.
• The XGBoost model, utilizing the selected biomarkers, offers a highly accurate and efficient diagnostic tool for breast cancer detection.