Application of the Gene Expression Omnibus (GEO) to Generate and Validate Consensus Transcriptomic Profiles that Accurately Differentiate Nodal Status in Colorectal Cancer

Background: Disparity persists between published differential gene expression profiles derived from colorectal cancer. The continued expansion of public gene expression repositories (PGER) such as Gene Expression Omnibus (GEO) as well as development of integrative techniques could now permit consensus profile development. Objective: The aim was to utilise PGER to generate and test consensus profiles differentiating early (non metastatic) and late (metastatic) stage colorectal cancer. Methods: PGER were searched for experiments examining non metastatic and metastatic colorectal cancer. An automated process was developed to generate four consensus transcriptomic profiles. The utility of these was tested in terms of differentiating node negative (stage 2) and node positive (stage 3) disease. Utility was further characterised using a stringent enrichment analysis. Results: Consensus profiles were generated on (a) concordant and (b) discordant dysregulation across the majority of experiments. Accuracy in differentiating stage two and three colorectal cancer ranged from 75 % to 97 % depending on data set and classification technique. Areas under the sumary reciever operating characteristic curves ranged from 0.73 to 0.86. Further validation confirmed associations with the colorectal metastatic process identifying established and novel therapeutic agents Conclusions: GEO based archives permit consensus profile generation. When subjected to rigorous bioinformatic processing these are highly accurate in discriminating nodal status in colorecatl cancer. Nodal status can now be determined based on transcriptomic profiles.