Detection and localization of gastrointestinal cancers based on multi-dimentional signatures from a single cfDNA targeted sequencing assay.

• DOI: 10.1200/jco.2023.41.16_suppl.4169
• Publication Date: 2023-06-04T15:37:26Z
• Authors: Xin-Rong Yang; Dong-Li He; Zhi-Guo Xiong; Bin Yan; Quan-Lin Li; De-Zhen Guo; Ao Huang; Zhen Feng; Pin-Xiang Lu; Qi Guo; Meng-Jiang He; Wei-Zhong Chen; Qi-Ye He; Zhi-Xi Su; Rui Liu; Yunshi Zhong; Jia Fan; Jian Zhou
• Keywords: Gastrointestinal cancer; Liquid biopsy; Cell-free fetal DNA

4169 Background: Five major gastrointestinal (GI) cancers - colorectal (CRC), gastric (GC), liver (LC), esophageal (EC), and pancreatic cancer (PC) are responsible for hundreds of thousands mortalities annually worldwide. Unfortunately, there is a lack cost-effective, blood-based screening method their early detection. To address this issue, we aimed to develop GutSeer, noninvasive, targeted methylation sequencing-based test by leveraging fragmentomic signatures carried cell-free DNA (cfDNA). Methods: The panel GutSeer consists 1656 target regions which were either differentially methylated between healthy samples, or distinctively in specific GI cancer. Cancer participants recruited randomly divided into training validation cohort. Their plasma samples analyzed generate features. These multi-dimensional features integrated build ensemble stacked machine learning models differentiate against healthy, determine the tissue-of-origin (TOO) Results: A total 1844 cases (787 342 LC, 239 GC, 209 EC, 180 CRC, 87 PC cases) study. cancer- vs-healthy model achieved an AUC 0.94 0.95 (sensitivity 77.7% 77.1% under specificity around 96%) using only, respectively. Combining both further improved performances, achieving 0.96 = 86.2% at 96.7%). For individual type cancer, has sensitivity 93.3% 81.1% 70.3% 96.5% 86.4% PC. An independent 629 benign as controls 87.1%. separate TOO was built all overall accuracy 82% (66.7% 87.0% GC EC combined, 89.0% 63.2% PC). Same detection model, prediction yielded higher than when only fragmentomics (accuracy 75.6% 75.4%, respectively). When compared with whole-genome sequencing (WGS) based approaches, showed comparable performance but identification, confirming its effectiveness cancers. Conclusions: non-invasive integrating features, demonstrated detect localize 5 main types high accuracy. Our results that reasonably sized can perform comparably even better WGS-based methods localization, indicating may be low-cost solution