A Neural Network–Enabled, Enzymatic cfDNA Methylation Assay for Colorectal Cancer Early Detection
Menée à partir de 216 échantillons plasmatiques prélevés sur 130 témoins en bonne santé et 86 patients atteints d'un cancer colorectal, cette étude évalue la sensibilité et la spécificité de modèles utilisant des réseaux de neurones et basés sur la méthylation de l'ADN libre circulant pour détecter précocement la maladie
Early detection of colorectal cancer (CRC) remains critical for reducing disease-specific mortality, yet current noninvasive screening approaches have limitations in sensitivity, patient adherence, and scalability. We developed and clinically evaluated a non–next-generation sequencing (non-NGS) liquid biopsy assay for CRC detection based on methylation profiling of circulating cell-free DNA (cfDNA). The assay focuses on 40 CpG regions selected via bioinformatic analysis of public methylome datasets and uses a TET2–APOBEC enzymatic conversion method to maintain cfDNA integrity and enhance amplification efficiency, enabling a rapid and cost-effective qPCR-based workflow. Methylation signals were quantified by qPCR and integrated with patient age using neural network–based predictive models. The assay was evaluated in a cohort of 216 plasma samples, including 86 CRC cases and 130 healthy controls. In the validation subset, 14 high-performing models demonstrated sensitivities ranging from 80.8% to 92.3% and specificities from 84.6% to 97.4%. A representative model achieved a validation sensitivity of 92.3% (95% CI, 75–99%), with early-stage (Stage I/II) sensitivity of 100 % (95% CI, 72–100%) at a specificity of 97.4% (95% CI, 87–100%). These findings support the potential of an enzymatic conversion–based, machine learning–guided cfDNA methylation assay as a practical, scalable, and minimally invasive approach for CRC detection. However, the relatively limited number of early-stage cases in this study highlights the need for larger, prospectively collected cohorts to refine performance estimates and confirm clinical utility.
Cancer Prevention Research , résumé, 2026