Colorectal cancer (CRC) is a high incidence disease, characterized by high morbidity and mortality rates. Early diagnosis remains challenging because fecal occult blood screening tests have performed sub-optimally, especially due to hemorrhoidal, inflammatory, and vascular diseases, while colonoscopy is invasive and requires a medical setting to be performed. The objective of the present study was to determine if serum metabolomic profiles could be used to develop a novel screening approach for colorectal cancer. Furthermore, the study evaluated the metabolic alterations associated with the disease. Untargeted serum metabolomic profiles were collected from 100 CRC subjects, 50 healthy controls, and 50 individuals with benign colorectal disease. Different machine learning models, as well as an ensemble model based on a voting scheme, were built to discern CRC patients from CTRLs. The ensemble model correctly classified all CRC and CTRL subjects (accuracy = 100%) using a random subset of the cohort as a test set. Relevant metabolites were examined in a metabolite-set enrichment analysis, revealing differences in patients and controls primarily associated with cell glucose metabolism. These results support a potential use of the metabolomic signature as a non-invasive screening tool for CRC. Moreover, metabolic pathway analysis can provide valuable information to enhance understanding of the pathophysiological mechanisms underlying cancer. Further studies with larger cohorts, including blind trials, could potentially validate the reported results.
Keywords: colorectal cancer; ensemble machine learning; fecal occult blood test; metabolomics; screening test.