Robust and stable gene selection via Maximum-Minimum Correntropy Criterion.

Area of Science:

• Bioinformatics and Computational Biology
• Cancer Research
• Genomics

Background:

• Identifying significant genes from thousands in microarray data is a major challenge in cancer research.
• Early detection and prevention of cancer progression rely on pinpointing the most involved genes.

Purpose of the Study:

• To propose a novel Maximum-Minimum Correntropy Criterion (MMCC) approach for effective and robust gene selection from microarray datasets.
• To determine the optimal number of features using an evolutionary optimization process.

Main Methods:

• Developed a Maximum-Minimum Correntropy Criterion (MMCC) algorithm for gene selection.
• Employed an evolutionary optimization process to identify the optimal number of features.
• Evaluated MMCC performance against established gene selection algorithms on 25 common microarray datasets.

Main Results:

• MMCC demonstrated superior performance compared to other gene selection algorithms across 25 microarray datasets.
• The MMCC approach is stable, fast, and robust against noise and outliers.
• High classification accuracy was achieved using Support Vector Machine (SVM) with fewer than 10 genes selected by MMCC.

Conclusions:

• The MMCC approach offers a highly accurate and efficient method for identifying informative genes in cancer research.
• MMCC significantly outperforms existing algorithms, providing a robust tool for gene selection in bioinformatics.
• This method facilitates the identification of a minimal yet highly predictive set of genes for cancer classification.