Abstract
OBJECTIVE: This study aimed to identify biomarkers for colorectal cancer (CRC) with representative gene functions and high classification accuracy in tissue and blood samples. METHODS: We integrated CRC DNA methylation profiles from The Cancer Genome Atlas and comorbidity patterns of CRC to select biomarker candidates. We clustered these candidates near the promoter regions into multiple functional groups based on their functional annotations. To validate the selected biomarkers, we applied 3 machine learning techniques to construct models and compare their prediction performances. RESULTS: The 10 screened genes showed significant methylation differences in both tissue and blood samples. Our test results showed that 3-gene combinations achieved outstanding classification performance. Selecting 3 representative biomarkers from different genetic functional clusters, the combination of ADHFE1, ADAMTS5, and MIR129-2 exhibited the best performance across the 3 prediction models, achieving a Matthews correlation coefficient > .85 and an F1-score of .9. CONCLUSIONS: Using integrated DNA methylation analysis, we identified 3 CRC-related biomarkers with remarkable classification performance. These biomarkers can be used to design a practical clinical toolkit for CRC diagnosis assistance and may also serve as candidate biomarkers for further clinical experiments through liquid biopsies.