The Metabolome and Transcriptome are mutually communicating within cancer cells, and this interplay is translated into the existence of quantifiable correlation structures between gene expression and metabolite abundance levels. Studying these correlations could provide a novel venue of understanding cancer and the discovery of novel biomarkers and pharmacological strategies, as well as laying the foundation for the prediction of metabolite quantities by leveraging information from the more widespread transcriptomics data. In the current paper, we investigate the correlation between gene expression and metabolite levels in the Cancer Cell Line Encyclopedia dataset, building a direct correlation network between the two molecular ensembles. We show that a metabolite/transcript correlation network can be used to predict metabolite levels in different samples and datasets, such as the NCI-60 cancer cell line dataset, both on a sample-by-sample basis and in differential contrasts. We also show that metabolite levels can be predicted in principle on any sample and dataset for which transcriptomics data are available, such as the Cancer Genome Atlas (TCGA).
Keywords: cancer; correlation networks; machine learning; metabolomics; transcriptomics.