The role of nonlinear dimension reduction of genome-wide DNA methylome in integrated diagnosis: A case study of glioblastoma, IDH-wildtype.

Abstract

To the Editor, Genome‐wide analysis of the DNA methylome has become one of the most important modalities for the integrated diagnosis of central nervous system (CNS) tumors, as demonstrated by the inclusion of this analysis in the fifth edition of the World Health Organization classification (WHO CNS5). Capper et al. showed that array‐based DNA methylome profiling can be used to classify CNS tumors using unsupervised machine‐learning techniques. Recent research supports the use of DNA methylation‐based classification, particularly in challenging cases involving pathological diagnoses. The DKFZ classifier, developed by the German Cancer Research Center (Deutsches Krebsforschungszentrum, DKFZ), has currently been considered the standard for DNA methylome classification. Raw intensity data (IDAT) files generated through the Illumina Infinium Methylation EPIC BeadChip arrays (Illumina Inc., San Diego, CA, USA), which provide genome‐wide DNA methylation profiles, were uploaded to a website (https:// www.molecularneuropathology.org/mnp/) to use the DKFZ classifier. The DKFZ classifier uses machine‐ learning techniques similar to artificial intelligence to perform unsupervised clustering and classify the methylation classes. The calibrated score produced by the classifier indicates the likelihood of a methylation class classification, with scores ranging between 0 and 1. Methylation class scores of ≥0.9 and <0.9 are considered match and unmatched, respectively. Importantly, the DKFZ classifier raises the awareness that DNA methylation data may still be relevant in unmatched cases with low tumor content or low DNA quality. Nonlinear dimensionality reduction techniques, such as t‐distributed stochastic neighbor embedding (t‐SNE) and uniform manifold approximation and projection, may be helpful via allowing the visualization of the similarity of the genome‐wide DNA methylome profile with existing data on reference cases accompanied with clinicopathological information to improve our ability to analyze cases where the DKFZ classifier shows “no match” or a methylation class with a low calibrated score. These techniques can reveal clusters and patterns in high‐dimensional data that may be difficult to detect using conventional linear dimensionality reduction methods, such as principal component analysis. In this paper, we present a case of an infiltrating high‐grade glioma in a younger patient, in which dimensionality reduction of the genome‐wide DNA methylome complemented the integrated diagnosis. The project was approved by an institutional ethics committee. The investigation was conducted in accordance with the Declaration of Helsinki of 1975. A 42‐year‐old Japanese man presented to a clinic with headache, nausea, dizziness, and visual disturbance that first appeared 2 months earlier. Brain magnetic resonance imaging revealed an approximately 6.0 cm in diameter T2/fluid‐attenuated inversion recovery (FLAIR) hyperintense (Supporting information: Figure 1) and gadolinium‐enhanced (Figure 1a) mass lesion with hemorrhage in the right temporal lobe of the cerebral hemisphere (Figure 1b). Gross total resection of the lesion was achieved at the Kyorin University Hospital, and a Stupp regimen consisting of radiotherapy (60 Gy/30 fractions) and temozolomide (120mg/m/day) was postoperatively administered. The patient returned to work 5 months after the surgery and was followed up for 7 months with a portable medical device (Optune) without enlargement of the contrast‐enhancing lesion. Histologically, atypical glial cells diffusely infiltrated the brain parenchyma. The tumor cells exhibited