Advancing Gene Feature Selection: A Synergistic Approach with Co-expression Networks and Genetic Algorithms

Gene feature selection is essential in bioinformatics and medical research, as it identifies gene subsets closely associated with specific diseases or biological traits from high-dimensional gene datasets. High-dimensional gene data causes the curse of dimensionality, leading to sparsity, complex inter-feature relationships, and significant noise. These issues undermine the reliability of conventional statistical methods in capturing underlying biological information. While gene feature selection can enhance classification model accuracy and reduce computational complexity, existing methods often struggle to handle the complexities of high-dimensional medical gene data.

To address the limitations of current methods, we designed a synergistic gene feature selection approach (CJWGA) that integrates co-expression networks and genetic algorithms. The goal is to efficiently perform gene feature selection, significantly reduce the size of feature subsets, and achieve high predictive accuracy across multiple gene datasets.

The CJWGA decomposes feature selection into two key steps: preprocessing of co-expression networks and iterative selection using genetic algorithms. A preprocessing gene selection approach (IMGCNet) is proposed to screen module genes based on conditional mutual information. For joint mutual information, a combined information entropy crossover operator (CIECO) and a joint adaptive mutation operator (JAMO) are designed for the nondominated sorting genetic algorithm, aiming to balance intensification and diversification.

Experimental results demonstrate that the proposed CJWGA achieves remarkable performance. It significantly reduces the size of feature subsets while maintaining high predictive accuracy across multiple gene datasets.

Overall, the synergistic CJWGA approach, integrating co-expression networks and improved genetic algorithms, exhibits excellent performance in gene feature selection. It addresses the challenges of high-dimensional medical gene data and holds potential as a valuable tool for gene feature selection in bioinformatics and medical research.