Population characteristic exploitation-based multi-orientation multi-objective gene selection for microarray data...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Gene selection from microarray data is crucial for effective cancer diagnosis and classification.
• Swarm intelligence algorithms often face premature convergence to local optima, hindering optimal gene subset identification.

Purpose of the Study:

• To propose a novel gene selection approach, MOMOGS-PCE, that reframes local optima as an opportunity rather than an obstacle.
• To enhance the discovery of globally optimal gene subsets for cancer diagnosis.

Main Methods:

• Developed MOMOGS-PCE, a multi-objective gene selection approach utilizing a 'reverse-thinking' strategy.
• Implemented a novel population initialization strategy with multiple diverse populations.
• Employed an enhanced NSGA-II algorithm to amplify population characteristics.
• Introduced a novel exchange strategy for inter-population characteristic transfer.

Main Results:

• MOMOGS-PCE demonstrated significant advantages in comprehensive indicators compared to six other multi-objective gene selection algorithms.
• The 'reverse-thinking' approach successfully avoided local optima and leveraged them for superior gene subset discovery.
• Validated the effectiveness of MOMOGS-PCE in identifying optimal gene subsets for cancer diagnosis.

Conclusions:

• The proposed MOMOGS-PCE approach offers a paradigm shift in gene selection by utilizing local optima.
• This method enhances the identification of discriminative gene subsets, leading to improved accuracy in cancer classification.
• MOMOGS-PCE provides a robust and effective tool for genomic data analysis in cancer research.