A Joint Analysis of Transcriptomic and Metabolomic Data Uncovers Increased Metabolic Transcriptional Regulation in Breast Cancer

Recent years have witnessed a growing interest in the study of cancer metabolism, facilitated by genome scale omics measurements that uncover the metabolic alterations occurring in cancer versus normal cells. Here we harness machine learning tools to analyze metabolomic and transcriptomic profiles jointly collected from breast cancer patients to identify metabolic reactions controlled by transcriptional regulation and construct predictors of metabolite levels from the much more ubiquitous gene expression measurements. Analyzing the existing measured data we find that cancer cells exhibit a greater number of significant transcriptional-metabolite associations than normal cells, testifying to an overall marked increase in the transcriptional regulation of cancer metabolism. Following, we show that we can predict these associations on unseen data with an accuracy exceeding 90%. Building on these results we further construct genome-wide predictors of metabolite levels from gene expression data, which correctly evaluate the levels of ~60% of the measured metabolites across samples based on their associated expression data alone. Finally, we apply our pipeline to a large cohort of breast cancer samples to predict their corresponding metabolite levels and examine their association with patients’ survival. We find that low levels of key known cancer related metabolites, including glucose, glycine, serine and acetate are significantly associated with improved survival time, and identify new predicted metabolites of interest. Overall, this analysis is the first to identify and chart a global increase in metabolic transcriptional regulation in cancer, and the first to provide means for inferring metabolite levels from transcriptional data in breast cancer on a genome wide level. We are currently working to extend our analysis to liver cancer (HCC) and pancreatic cancer (PDAC).