| Timetable |
| Download Conference Program |
| Download All Abstracts |
| Zoom Access |
| Corporate Program |
Poster Presentations
Day 3, June 24（Tue.）
Room P (Maesato East, Foyer, Ocean Wing)
- 3P-PM-05
Integrating Untargeted and Targeted Metabolomics with Machine Learning for Early Colorectal Cancer Biomarker Discovery
(1NTOU, 2NCKU)
oPang-Hung Hsu1, Chung-Fa Chang2, Chung-Hsien Lin2, Juan-Kai Wong2
Colorectal cancer (CRC) poses a notable global health challenge, necessitating improved diagnostic and therapeutic strategies. This study developed an innovative multi-platform metabolomics workflow that combines targeted and untargeted approaches to identify novel CRC biomarkers. First, by analyzing high-resolution mass spectrometry data extracted from the serum samples of CRC patients and healthy controls using an untargeted random forest algorithm for feature selection and subsequent database annotation, five key metabolites were identified: N-methylcytisine, 2-piperidone, theophylline, DL-norleucine, and linolenic acid. These metabolites consistently exhibited lower concentrations in CRC patients compared to healthy individuals across both discovery and validation cohorts. Cell-based assays revealed that, while these metabolites did not considerably affect cancer cell proliferation, most of them demonstrated strong inhibitory effects on the migration and invasive capabilities of CRC cell lines. A multimarker panel incorporating these metabolites demonstrated improved predictive capability for CRC detection, achieving 93.7% accuracy, 97.9% sensitivity, and 89.4% specificity. Notably, three of the five metabolites (2-piperidone, norleucine, and linolenic acid) are associated with gut bacteria, suggesting a potential link between the gut microbiome and CRC pathogenesis. Overall, this study exemplifies an innovative strategy for precision cancer diagnostics and therapeutics, paving the way for personalized medicine in CRC management.