Abstract
The accumulation of cancer metabolomics data in the past decade provides exceptional opportunities for deeper investigations into cancer metabolism. However, integrating a large amount of heterogeneous metabolomics data to draw a full picture of the metabolic reprogramming and to discover oncometabolites of certain cancers remains challenging. In this study, a tumor barcode constructed based upon existing metabolomics "big data" using the Bayesian vote-counting method is proposed to identify oncometabolites in colorectal cancer (CRC). Specifically, a panel of oncometabolites of CRC was generated from 39 clinical studies with 3202 blood samples (1332 CRC vs. 1870 controls) and 990 tissue samples (495 CRC vs. 495 controls). Next, an oncometabolite-protein network was constructed by combining the tumor barcode and its involved proteins/enzymes. The effect of anti-cancer drugs or drug combinations was then mapped into this network by the random walk with restart process. Utilizing this network, potential Irinotecan (CPT-11)-sensitizing agents for CRC treatment were discovered by random forest and Xgboost. Finally, a compound named MK-2206 was highlighted and its synergy with CPT-11 was validated on two CRC cell lines. To summarize, we demonstrate in the present study that the metabolomics "big data"-based tumor barcodes and the subsequent network analyses are potentially useful for drug combination discovery or drug repositioning.