Abstract 23P
Background
Among various genomic features in the blood, methylation pattern could be useful not only in determining whether cancer is present, but in revealing its origin. The purposes of this prospective study were to evaluate the diagnostic performance of methylation sensitive restriction enzyme digestion followed by sequencing method (MRE-seq) using cell-free DNA (cfDNA) in the diagnosis of cancer and to investigate tissue-of-origin (TOO) using deep neural network (DNN).
Methods
We have developed a deep learning-based prediction model with the hypomethylation data obtained from MRE-seq. One hundred ninety-one cancer patients with stages I–IV (95 lung cancer, 96 colorectal cancer) and 126 non-cancer participants were enrolled in this study. DNA isolated from each sample was subjected to MRE-seq to analyze the difference in methylation patterns between cancer tissues and normal tissues.
Results
We achieved the overall area under the ROC curve (AUC) 0.978 with sensitivity 78.1% at specificity 99.2% for colorectal cancer, AUC 0.956 with sensitivity 66.3% at specificity 99.2% for lung cancer. In colorectal cancer, the sensitivity is 76.5% in stage I, 76.2% in stage II, 78.3% in stage III and 83.3% in stage IV at specificity 99.2%. In lung cancer, the sensitivity is 48.5% in stage I, 44.4% in stage II, 78.9% in stage III and 76.0%enl in stage IV at specificity 99.2 %. The true positive rate (TPR) of TOO model was 94.4% (85/90), 89.9% (80/89) for colorectal cancer and lung cancer respectively.
Conclusions
Global hypomethylation pattern obtained by MRE-seq was useful in the diagnosis of cancer with high accuracy and in the determination of TOO through DNN. The role of MRE-seq needs to be validate for various solid tumors in the future multicenter studies.
Clinical trial identification
NCT 04253509.
Editorial acknowledgement
Legal entity responsible for the study
Eone Diagnomics Genome Center, Republic of Korea.
Funding
Has not received any funding.
Disclosure
All authors have declared no conflicts of interest.