Dual-Population Multi-Objective Optimization With Multi-Scale Co-Expression Modeling for Medical Gene Expression...

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning

Background:

• High-dimensional gene expression data presents challenges for feature selection due to redundancy and instability.
• Existing methods struggle with the trade-off between predictive accuracy and the number of selected features.

Purpose of the Study:

• To develop a robust two-stage framework, CJHMO, for effective medical gene expression feature selection.
• To enhance the performance-compression trade-off in feature selection for high-dimensional datasets.

Main Methods:

• Stage one employs a Multi-Scale Co-Expression Attention Network (MSCANet) for structural candidate generation and module summarization.
• Stage two utilizes a Dual-Population Heterogeneous Multi-Objective optimizer (DPHMO) for wrapper-based optimization, combining global exploration and local refinement.
• An external elite archive facilitates information exchange between populations.

Main Results:

• CJHMO demonstrates superior performance-compression trade-offs compared to existing multi-objective methods.
• Experiments show improved Pareto quality, indicated by higher Hypervolume (HV) and lower Indicator of Generational Distance (IGD) values.
• The framework effectively addresses challenges in high-dimensional, small-sample gene expression data.

Conclusions:

• The proposed CJHMO framework offers a significant advancement in medical gene expression feature selection.
• It provides a better balance between predictive performance and feature subset compactness.
• CJHMO shows promise for applications requiring efficient and accurate gene selection in complex biological data.