A Feature Selection Method Based on Graph Theory for Cancer Classification.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Gene expression profile data is valuable for tumor research.
• High dimensionality and redundancy in gene expression data pose challenges for classification.
• Effective gene selection is critical for accurate microarray data analysis.

Purpose of the Study:

• To develop an effective gene selection method for microarray data classification.
• To address the challenges of high dimensionality and redundancy in gene expression data.
• To improve the accuracy and robustness of tumor classification using gene expression profiles.

Main Methods:

• A novel feature selection method integrating maximum mutual information coefficient and graph theory.
• Representing gene expression data features as vertices in an undirected graph.
• Utilizing core anditivity theory to identify informative gene subsets.

Main Results:

• The proposed algorithm demonstrated high accuracy and robustness across six diverse genetic datasets.
• Performance was evaluated using multiple classification models (accuracy, F1-Score, AUC) to ensure reliability.
• The method outperformed existing advanced feature selection techniques.

Conclusions:

• The developed method simultaneously considers feature importance and correlation.
• This approach effectively solves the gene selection problem in microarray data classification.
• The findings contribute to more accurate and reliable tumor classification from gene expression data.