EpiGe: A machine-learning strategy for rapid classification of medulloblastoma using PCR-based methyl-genotyping

Abstract

Molecular classification of medulloblastoma is critical for the treatment of this brain tumor. Array-based DNA methylation profiling has emerged as a powerful approach for brain tumor classification. However, this technology is currently not widely available. We present a machine-learning decision support system (DSS) that enables the classification of the principal molecular groups-WNT, SHH, and non-WNT/non-SHH-directly from quantitative PCR (qPCR) data. We propose a framework where the developed DSS appears as a user-friendly web-application-EpiGe-App-that enables automated interpretation of qPCR methylation data and subsequent molecular group prediction. The basis of our classification strategy is a previously validated six-cytosine signature with subgroup-specific methylation profiles. This reduced set of markers enabled us to develop a methyl-genotyping assay capable of determining the methylation status of cytosines using qPCR instruments. This study provides a comprehensive approach for rapid classification of clinically relevant medulloblastoma groups, using readily accessible equipment and an easy-to-use web-application.t.

Keywords: Cancer; Health technology; Machine learning.

Graphical abstract

Figure 1

Patient flow diagram CNS, central nervous system; DSS, decision support system; EPIC, Illumina methylation EPIC BeadChip array; HM450K, Illumina Infinium HumanMethylation 450 BeadChip; MB, medulloblastoma; qPCR, quantitative PCR. Images created with BioRender.

Figure 2

Development and testing of the methyl-genotyping assays (A) The decision support system (DSS) was generated using 38 medulloblastoma cases (training cohort) with available methyl-genotyping qPCR data (EpiGe) and methylation microarray data (EPIC, Illumina methylation EPIC BeadChip array) of the six-cytosine signature. (B) Bisulfite conversion (BS) scheme. (C) The mean DNA methylation values of the six-cytosine signature that accurately discriminate the medulloblastoma subgroups WNT, SHH, and non-WNT/non-SHH (Gómez et al. 2018). (D) The quantitative PCR (qPCR) amplification curve representation of bisulfite-converted cytosines for methylated cytosines (with FAM labeling) in red, and unmethylated cytosines (with VIC labeling), in blue.

Figure 3

DNA methylation status predictor (A) Allele 1 and 2 ΔRn correlation between replicates. Pearson correlation coefficient (R²) is 0.98 and 0.95 for ΔRn Allele 1 and 2, respectively; p value <0.01 for both correlations. (B) Left, Allele 1 ΔRn mean value database in a base 10 logarithmic scale. Right, the Allele 2 ΔRn mean value database in a base 10 logarithmic scale. (C) Receiver operating characteristic (ROC) curve of the logistic regression method (LRM). TPR, true positive rate; FPR, false positive rate. (D) Donut pie of previously published genome-wide DNA methylation array data of medulloblastoma (MB) tumors (n = 3,044) The outer pie represents the distribution of the binarized methylation status. (E) Donut pie of non-MB tumors analyzed by genome-wide DNA methylation array data of (n = 1,644). (F). HD score radar plots of three prototypical binary codes, of perfect match (HD = 1) with no mismatches, 1 mismatch (HD = 0.83), or 2 mismatches (the maximum number accepted) (HD = 0.67). (G) Confusion matrix of the Hamming distance (HD) prediction of the training cohort analyzed by EpiGe. All sample replicates were assigned to the nearest MB subgroup reference binary code (94.74% [95% CI, 82.25%–99.36%]; k = 0.87).