Integrative analysis of plasma cell-free DNA fragmentation and methylation patterns for colorectal cancer detection.

Abstract

e15022 Background: Recently, the analysis of epigenetic traits of cfDNA has been ben proven to be effective in early non-invasive cancer detection. Although integrating the DNA fragmentation and methylation information is expected to improve cancer detection performances, there are technical hurdles for effective integration of these data for detecting cancer non-invasively. In this study, we generated and analyzed targeted methylation sequencing data to develop a cancer detection method that uses both methylation and DNA fragment size patterns, optimized parameters for the model and evaluated the accuracy of the model in colon adenocarcinoma patients. Methods: Plasma cfDNA was collected from a total of 215 donors (111 COAD patients and 104 healthy adults) and targeted methylation sequencing data was generated using a customized panel covering 196k CpG sites containing robustly evaluated cancer specific markers. A total of 60 COAD and healthy samples, respectively, were used to train the model using elastic net and logistic regression. The remaining COAD 51 and healthy 44 samples were used as the test dataset in evaluating the model. Results: The integrative method achieved a high performance (94.12% sensitivity and 97.93% specificity) in classifying COAD patients and healthy individuals. Interestingly, when we decreased the bin size for cfDNA fragment length analysis, the performance of the model was significantly improved even with the same number of bins. Since the model with smaller bin size performed better, we could decrease the size target regions and data size. Our results proposed that analyzing fragment size at high resolution was an effective way to use the feature for cancer detection. By designing targeted Methyl-seq panel and generating high depth sequencing data (> 1000X), we could effectively integrate DNA methylation and fragmentation patterns for non-invasive cancer detection. Conclusions: In this study, we propose a novel non-invasive CRC detection method using targeted Methyl-seq data based on the integration of DNA methylation and fragmentation patterns. The adding of fragmentation features to the methylation information was effective in improving the sensitivity and specificity of CRC detection.