Related Experiment Video
Updated: Jul 26, 2025
Genome-Wide Analysis of DNA Methylation in Gastrointestinal Cancer
Published on: September 18, 2020
DNA methylation-based classifier differentiates intrahepatic pancreato-biliary tumours
Mihnea P Dragomir1, Teodor G Calina2, Eilís Perez3
1Institute of Pathology, Charité - Universitätsmedizin Berlin, Corporate Member of Freie Universität Berlin, Humboldt-Universität zu Berlin, Berlin, Germany; German Cancer Consortium (DKTK), Partner Site Berlin, and German Cancer Research Center (DKFZ), Heidelberg, Germany; Berlin Institute of Health, Berlin, Germany.
Machine learning models accurately distinguish intrahepatic cholangiocarcinomas (iCCA) from pancreatic ductal adenocarcinoma (PAAD) liver metastases using DNA methylation data. This DNA methylation classifier improves pancreato-biliary cancer diagnosis.
Area of Science:
- Oncology
- Genomics
- Bioinformatics
Background:
- Distinguishing intrahepatic cholangiocarcinoma (iCCA) from hepatic metastases of pancreatic ductal adenocarcinoma (PAAD) is diagnostically challenging due to similar morphology and shared mutations.
- Current diagnostic methods often struggle to differentiate these two distinct liver cancers.
- Hypothesis: DNA methylation patterns analyzed by machine learning can effectively differentiate iCCA from PAAD liver metastases.
Purpose of the Study:
- To develop and validate a DNA methylation-based machine learning classifier for differentiating iCCA from PAAD liver metastases.
- To assess the accuracy of machine learning models in classifying pancreato-biliary cancers of the liver.
- To provide a tool for improving the diagnostic accuracy of these challenging liver cancers.
Main Methods:
- Genome-wide DNA methylation data from iCCA (n=259), PAAD (n=431), and normal bile duct (n=70) tissues were compiled from public datasets.
- Data were split into reference (n=399) and validation (n=361) sets; three machine learning models (neural network, support vector machine, random forest) were trained on the reference set.
- Classifiers were validated on the technical validation set and further tested on an internal cohort (n=72).
Main Results:
- On the validation cohort, neural network, support vector machine, and random forest classifiers achieved accuracies of 97.68%, 95.62%, and 96.5%, respectively.
- Filtering by anomaly detection improved accuracies to 99.07% (neural network), 96.22% (support vector machine), and 100% (random forest).
- The neural network classifier, with applied filters, achieved 95.45% accuracy on an independent internal cohort.
Conclusions:
- A highly accurate DNA methylation-based classifier was developed to differentiate iCCA, PAAD liver metastases, and normal bile duct tissue.
- The developed tool demonstrates significant potential for enhancing the diagnostic process of pancreato-biliary cancers affecting the liver.
- Machine learning analysis of DNA methylation provides a robust method for distinguishing between morphologically similar liver tumors.
