Motivation: Large-scale gene expression analysis is a valuable asset for data-driven hypothesis generation. However, the convoluted nature of large expression datasets often hinders extraction of meaningful biological information. Results: To this end, we developed GECO, a gene expression correlation analysis software that uses a genetic algorithm-driven approach to deconvolute complex expression datasets into two subpopulations that display positive and negative correlations between a pair of queried genes. GECO's mutational enrichment and pairwise drug sensitivity analyses functions that follow the deconvolution step may help to identify the mutational factors that drive the gene expression correlation in the generated subpopulations and their differential drug vulnerabilities. Finally, GECO's drug sensitivity screen function can be used to identify drugs that differentially affect the subpopulations. Availability and implementation: http://www.proteinguru.com/geco/ and http://www.proteinguru.com/geco/codes/. Supplementary information: Supplementary data are available at Bioinformatics online.