Yin Huihui, Liu Xin, Chen Yu, Huang Jie, Li Lili
Objective: To explore the potential application of DNA methylation detection based on next-generation sequencing (NGS) for early screening of solid tumors, and to analyze the standardization issues in assay development, quality control, and performance evaluation, thereby providing references for future clinical translation. Methods: Key steps of methylation sequencing in tumor early screening were systematically reviewed, including cancer type selection and biomarker discovery, methylation conversion approaches (chemical and enzymatic methods), library preparation strategies (single-strand and double-strand), sequencing platform differences (Illumina vs. BGI DNBSEQ), and calculation approaches for methylation signals. The application and performance of various modeling methods (logistic regression, decision tree, random forest, gradient boosting tree, etc.) were compared, and the current practices and challenges in quality control and clinical validation were summarized, focusing on reference material preparation, plasma matrix evaluation, and performance metrics. Results: In library preparation, chemical conversion offers lower cost but causes severe DNA degradation, whereas enzymatic conversion is milder but technically more complex. Single-strand libraries provide higher fidelity, while double-strand libraries are more established and scalable. Different sequencing platforms exhibit systematic differences in GC-region coverage and methylation signal measurement. The computational strategies and algorithms used for biomarker modeling significantly affect model performance, with ensemble learning methods (random forest, gradient boosting tree, etc.) generally achieving higher sensitivity and specificity than traditional models (logistic regression, decision tree, etc.). Consequently, ensemble models are often applied in large-scale clinical cohorts and product development, while traditional models remain valuable for mechanism exploration and biomarker selection due to their interpretability. Reference materials prepared from cell line DNA provide consistency and scalability but require fragmentation and gradient design to mimic plasma cfDNA. Plasma matrix interference from residual human DNA may bias methylation signals and represents a critical issue for quality control. Moreover, the limited accuracy of early cancer detection reduces the reliability of positive results, potentially leading to unnecessary medical interventions and underscoring the necessity of clinical validation. Conclusion: NGS-based methylation detection offers a promising approach for early cancer screening, with advantages in multi-target detection and tissue-of-origin analysis. However, its clinical implementation relies on the establishment of standardized workflows, robust quality control systems, and comprehensive performance evaluation frameworks. Future efforts should focus on developing reference standards, optimizing quality control checkpoints, and advancing clinical validation studies. These steps are crucial to assessing the feasibility and standardization of multi-cancer early detection in clinical practice.
