An Efficient Feature Selection Strategy Based on Multiple Support Vector Machine Technology with Gene Expression...

Area of Science:

• Bioinformatics
• Computational Biology
• Cancer Genomics

Background:

• Gene expression data analysis is crucial for cancer diagnosis and classification.
• High-dimensional gene expression data presents challenges in identifying relevant biomarkers.
• Accurate feature selection is essential for developing robust cancer classification models.

Purpose of the Study:

• To propose a novel feature selection method, SVM-RFE-PO, for identifying determinant genes in breast cancer.
• To evaluate the performance of different parameter optimization algorithms (GS, PSO, GA) within the SVM-RFE framework.
• To compare the proposed method against existing feature selection techniques for breast cancer classification.

Main Methods:

• Developed SVM-RFE-PO integrating Grid Search (GS), Particle Swarm Optimization (PSO), and Genetic Algorithm (GA) for optimal parameter tuning.
• Applied three variants: SVM-RFE-GS, SVM-RFE-PSO, and SVM-RFE-GA for feature selection.
• Utilized selected feature subsets to train SVM classifiers for cancer classification.
• Compared performance against Random Forest Feature Selection (RFFS) and Minimal Redundancy Maximal Relevance (mRMR).

Main Results:

• The SVM-RFE-PSO algorithm yielded a feature subset with superior prediction performance, indicated by a higher Area Under Curve (AUC) on the testing dataset.
• The proposed SVM-RFE-PO method, particularly SVM-RFE-PSO, proved to be time-saving.
• The method successfully extracted more representative and useful genes for breast cancer classification.

Conclusions:

• SVM-RFE-PSO is an effective and efficient method for selecting informative genes from high-dimensional breast cancer expression data.
• This approach enhances the accuracy of cancer classification models.
• The study highlights the potential of optimized feature selection techniques in cancer genomics research.