DNA methylation-based classification of hematolymphoid neoplasms

Accurate pathological diagnosis of hematolymphoid neoplasms (HLN) is often challenging due to their complexity and heterogeneity. Genome-wide DNA methylation profiling has emerged as a valuable tool for tumor classification and diagnosis across malignancies, like central nervous system neoplasms. In this study, we explored the role of DNA methylation-based profiling in HLN. We generated the largest cross-platform HLN methylome cohort to date (1,156 samples) and identified 44 reproducible methylation classes (MCs) that aligned closely with WHO 5th edition/ICC entities, including subgroups with clinical and biological relevance. Copy number alterations were also inferred for all MCs. Additionally, a machine learning-based DNA methylation classifier was developed and validated on independent test set, demonstrating a modest 58% high-confidence score rate, nevertheless with a robust 97% concordance with the original diagnosis in these high-confidence score cases. Additionally, in discrepant high confidence score cases, while few, the methylation classifier demonstrated its potential utility as an adjunct where on additional review the diagnosis was revised in favor of the methylation prediction in the majority of cases (5/8 discrepant cases). Tumor purity was a significant contributor for a substantial proportion of low confidence score samples (scores below predetermined thresholds ,42%), affecting the classifier performance (X2 = 11.7, p0.0008). Our findings suggest that specific hematolymphoid tumor types exhibit distinct methylation signatures that can be leveraged to accurately classify these tumors. As a pilot study, our results provide a foundation for the development of a comprehensive and clinically valuable methylation-based classifier for hematolymphoid tumors in the future.