Abstract 1945: High-sensitive gastrointestinal cancers early detection by cell-free DNA targeted methylation sequencing with an innovative modeling approach

Abstract Background: Gastrointestinal (GI) cancers, including esophageal squamous cell carcinoma (ESCC), colorectal cancer (CRC) and stomach adenocarcinoma (STAD), contribute to nearly one-third of cancer-related deaths in China. The late-stage diagnosis of these cancers is largely due to limited access to endoscopic procedures and low screening adherence. A blood-based multi-cancer early detection (MCED) test offers a more convenient alternative; however, the efficacy in detecting early-stage cancers remains unsatisfactory. This study presents an MCED assay utilizing targeted DNA methylation sequencing to detect GI cancers by implementing an innovative modeling approach integrating synthesized early-stage cancer samples. Methods: A targeted sequencing panel was developed to identify tumor-specific hyper- and hypo-methylation markers. A cohort consisting of 1, 618 plasma samples (238 ESCC, 262 CRC, 309 STAD, 809 non-cancers controls; stage: I 15%, II 29%, III 40%, IV 16%) was collected from a prospective, multicenter study (the PROFUTURE study, NCT05874648) and randomly split into training and validation sets (4:1) with 20 iterations of splitting. A novel Methylation-Guided Synthetic Sample (MGSS) model was trained through a combination of real and synthetic samples using a machine learning XGBoost algorithm, validated by real samples. In each iteration, synthetic samples were generated by blending late-stage cancer plasma samples with a pool of 20 independent healthy plasma samples, optimized to closely mimic authentic early-stage methylation patterns. And the number of synthetic samples was set to half the number of cancer samples in the training set. Results: The MGSS model demonstrated high performance, achieving an area under the curve (AUC) of 0.934 (95% CI: 0.912-0.956), with a sensitivity of 69.7% (95% CI: 64.0%-75.5%) at a specificity of 98.8% (95% CI: 96.4%-100%). CRC exhibited the highest sensitivity at 83.8% (I 47.9%; II 81.4%, III 88.7%, IV 100.0%), followed by ESCC with a sensitivity of 77.6% (I 36.9%; II 67.6%; III 95.0%; IV 100.0%), and STAD at 53.7% (I 21.1%; II 46.9%, III 69.2%, IV 90.3%). Additionally, the sensitivity for stage I-III GI cancers was 68.1% (CRC: 82.8%; ESCC: 76.6%; STAD: 50.5%). Conclusions: The MGSS model, which leverages synthetic sample methodologies, represents a cost-effective advancement and performance improvement in GI cancers. This innovative approach holds promise for enhancing the performance of MCED for other cancers potentially improving early diagnosis and outcomes. Citation Format: Ping Lan, Zhihua Liu, Hong Xu, Xiaosheng He, Zhigang Li, Lei Lian, Yin Li, Songbing He, Xiaofeng Guo, Qun Zhao, Jianhong Lian, Zhijian Xu, Nan Lin, Jiandong Tai, Quan Wang, Feng Li, Jian Li, Lulu Zheng, Zhihui Xu, Yanzhan Yang, Yonghui Li, Baoliang Zhu, Xiaohui Wu. High-sensitive gastrointestinal cancers early detection by cell-free DNA targeted methylation sequencing with an innovative modeling approach [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 1945.