Weighted gene co-expression network analysis (WGCNA) is a widely used software tool that is used to establish relationships between phenotypic traits and gene expression data. It generates gene modules and then correlates their first principal component to phenotypic traits, proposing a functional relationship expressed by the correlation coefficient. However, gene modules often contain thousands of genes of different functional backgrounds. Here, we developed a stochastic optimization algorithm, known as genetic algorithm (GA), optimizing the trait to gene module relationship by gradually increasing the correlation between the trait and a subset of genes of the gene module. We exemplified the GA on a Japanese plum hormone profile and an RNA-seq dataset. The correlation between the subset of module genes and the trait increased, whereas the number of correlated genes became sufficiently small, allowing for their individual assessment. Gene ontology (GO) term enrichment analysis of the gene sets identified by the GA showed an increase in specificity of the GO terms associated with fruit hormone balance as compared with the GO enrichment analysis of the gene modules generated by WGCNA and other methods.