A novel multi-objective optimization framework using NSGA-II for gene co-expression network inference.

Area of Science:

• Bioinformatics
• Computational Biology
• Systems Biology

Background:

• Gene co-expression networks (GCNs) are crucial for understanding gene function and pathways from transcriptomic data.
• Noisy biological data often leads to unreliable GCNs with spurious connections and unrealistic structures.
• Existing methods struggle to balance network properties like sparsity and modularity effectively.

Purpose of the Study:

• To introduce a novel multi-objective optimization framework for refining edge selection in GCNs.
• To enhance the reliability and biological plausibility of GCNs derived from transcriptomic data.
• To simultaneously optimize multiple network characteristics, including sparsity, modularity, and scale-free topology.

Main Methods:

• Utilized Variance Stabilizing Transformation (VST) for RNA-seq data normalization.
• Employed Spearman rank correlation for robust co-expression estimation.
• Integrated Non-dominated Sorting Genetic Algorithm II (NSGA-II) for multi-objective network optimization.
• Incorporated permutation testing and bootstrap resampling for significance and stability assessment.

Main Results:

• The proposed NSGA-II based approach generated sparser and more modular GCNs compared to WGCNA and ARACNE.
• The optimized networks exhibited improved adherence to scale-free network properties across both microarray and RNA-seq datasets.
• The method demonstrated robust performance on heterogeneous transcriptomic datasets (GSE10245 and GSE102349).

Conclusions:

• The optimization-driven strategy provides a robust method for constructing high-quality GCNs.
• This approach offers a significant advancement for integrative genomic studies and biomarker discovery.
• The framework holds potential for improving the modeling of complex disease mechanisms.