Pan-Cancer Detection Through DNA Methylation Profiling Using Enzymatic Conversion Library Preparation with Targeted Sequencing

We investigated differences in circulating cell-free DNA (cfDNA) methylation between patients with cancer and those presenting with severe, nonspecific symptoms. Plasma cfDNA from 229 patients was analyzed, of whom 37 were diagnosed with a wide spectrum of cancer types within 12 months. Samples underwent enzymatic conversion, library preparation, and enrichment using the NEBNext workflow and Twist pan-cancer methylation panel, followed by sequencing. Methylation analysis was performed with nf-core/methylseq. Differentially methylated regions (DMRs) were identified with DMRichR. Machine learning with cross-validation was used to classify cancer and controls. The classifier was applied to an external validation set of 144 controls previously unseen by the model. Cancer samples showed higher overall CpG methylation than controls (1.82% vs. 1.34%, <i>p</i> < 0.001). A total of 162 DMRs were detected, 95.7% being hypermethylated in cancer. Machine learning identified 20 key DMRs for classification between cancer and controls. The final model achieved an AUC of 0.88 (83.8% sensitivity, 83.8% specificity), while mean cross-validation performance reached an AUC of 0.73 (57.1% sensitivity, 77.5% specificity). The specificity of the classifier on unseen control samples was 79.2%. Distinct methylation differences and DMR-based classification support cfDNA methylation as a robust biomarker for cancer detection in patients with confounding conditions.