Classification of central nervous system tumors based on DNA methylation expression spectrum

Ref: Capper D,et alNature. 2018 Mar 22;555 (7697):469-474doi: 10.1038/nature26000Epub 2018 Mar 14But tumor methylation testing can reflect changes in DNA methylation and cell origin in somatic cellsBecause of the high stability and repeatability of DNA methylation expression spectrum, it has been used in the subtype diagnosis of a variety of central nervous system tumorsDavid Capper, of Neuropathology at the University Hospital of Heidelberg, Germany, and others proposed a comprehensive classification of central nervous system tumors based on DNA methylation, published in Nature in March 2018the study included a total of 2801 cases containing 91 methylation classificationsThe authors first used illumina 450K methylation chip to test the whole genome methylation expression spectrum and used the t-SNE method to show the status of all cases after the reduction of dimensions (Figure 1)Figure 1DNA methylation expressionis clinically accurate and rapid diagnosis of tumor classification, the authors used random forest method to make a decision tree analysis of 2801 tumor samples of 91 methylation classificationsThe random forest method assigns each case to the methylation classification and produces an original score, which is then calibrated using logistic regressionThe results of the cross-validation show that the error rates of the original and calibration scores were 4.89 percent and 4.28 percent, respectivelyThe ROC curve shows 99% accuracy (Figure 2)Figure 2 Build and cross-verify DNA methylation classification further clinical feasibility assessment, the authors simultaneously examined 1155 cases for methylation expression testing and tissue pathology Eighty-eight percent of the test samples could be matched into the DNA methylation classification group, with a calibration classification value greater than 0.9 (Figure 3), of which 838 (76%) matched the results of histopathological examinations Figure 3 DNA methylation classification is used in clinical practice for assisted diagnosis 139 cases with calibration scores greater than 0.9, but their methylation classification did not match histopathological classification, the authors re-performed additional molecular tests, such as DNA copy numbers, targeted gene sequencing and fusion gene analysis, and corrected 129 of the 139 cases for histopathological diagnosis, which further supported DNA methylation classification (Figure 4) Reassessment of cases that do not correspond to histopathological diagnosis for samples that could not be classified as DNA methylation, the authors used unsupervised t-SNE analysis and found that the results were highly coincident with DNA methylation classification Most of the unclassified samples were clustered around or independently in the DNA methylation category; Figure 5 New types of central nervous system tumors can be discovered by DNA methylation classification.