A balanced iterative random forest for gene selection from microarray data.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• High-throughput microarray technologies generate complex, high-dimensional gene expression datasets.
• Imbalanced class distribution in biological data poses challenges for biomarker discovery.
• Identifying informative genes is crucial for disease diagnosis and understanding.

Purpose of the Study:

• Introduce the Balanced Iterative Random Forest (BIRF) algorithm.
• Select relevant genes from imbalanced high-throughput gene expression microarray data.
• Validate the selected genes as reliable biomarkers.

Main Methods:

• Application of the BIRF algorithm on four cancer microarray datasets.
• Comparison of BIRF performance against Support Vector Machine-Recursive Feature Elimination (SVM-RFE), Multi-class SVM-RFE (MSVM-RFE), Random Forest (RF), and Naive Bayes (NB).
• Validation of selected informative biomarkers through repeated training experiments.

Main Results:

• BIRF outperforms state-of-the-art methods, particularly on imbalanced datasets.
• Achieved 7%-12% higher accuracy than MSVM-RFE on a childhood leukaemia dataset, improving prediction for the minor class.
• 64% of top genes consistently appeared across validation experiments, indicating robust biomarker selection.

Conclusions:

• The BIRF algorithm is effective for gene selection from imbalanced high-throughput gene expression data.
• BIRF demonstrates superior performance compared to existing methods, especially in handling class imbalance.
• BIRF facilitates distinguishing truly predictive genes from those that appear predictive by chance.