Discrete Biogeography Based Optimization for Feature Selection in Molecular Signatures

Area of Science:

• Bioinformatics
• Computational Biology
• Cancer Research

Background:

• High-dimensional gene expression data present challenges in biomarker discovery due to redundancy and noise.
• Identifying a subset of informative genes is crucial for accurate cancer diagnosis and classification.
• Existing feature selection methods may not optimally handle the complexity of genomic datasets.

Purpose of the Study:

• To propose a novel discrete biogeography-based optimization (DBBO) algorithm for efficient gene selection.
• To enhance the accuracy and efficiency of biomarker discovery in cancer research.
• To evaluate the performance of DBBO against established optimization algorithms.

Main Methods:

• A Fisher-Markov selector was employed for initial gene data selection.
• Discrete migration and mutation models were developed to optimize biogeography-based optimization for feature selection.
• The proposed Discrete Biogeography-Based Optimization (DBBO) algorithm was implemented.
• DBBO was integrated with three classifiers and evaluated using 10-fold cross-validation on breast cancer datasets.

Main Results:

• The DBBO algorithm demonstrated effective gene subset selection for cancer classification.
• Experimental results on four breast cancer datasets showed DBBO's superior or comparable performance.
• DBBO outperformed or matched genetic algorithm, particle swarm optimization, differential evolution, and hybrid biogeography-based optimization.

Conclusions:

• The proposed DBBO method is an effective and efficient approach for feature selection in high-dimensional gene expression data.
• DBBO shows significant potential for improving biomarker discovery and cancer classification.
• This algorithm offers a robust solution for identifying discriminative genes relevant to different sample classes.