Guido Sauter

Jan 17, 2025 | nature.com | Yunhee Jeong |Clarissa Gerhauser |Guido Sauter |Thorsten Schlomm |Karl Rohr |Pavlo Lutsik

DNA methylation (DNAm) is a key epigenetic mark that shows profound alterations in cancer. Read-level methylomes enable more in-depth analyses, due to their broad genomic coverage and preservation of rare cell-type signals, compared to summarized data such as 450K/EPIC microarrays. Here, we propose MethylBERT, a Transformer-based model for read-level methylation pattern classification. MethylBERT identifies tumour-derived sequence reads based on their methylation patterns and local genomic sequence, and estimates tumour cell fractions within bulk samples. In our evaluation, MethylBERT outperforms existing deconvolution methods and demonstrates high accuracy regardless of methylation pattern complexity, read length and read coverage. Moreover, we show its applicability to cell-type deconvolution as well as non-invasive early cancer diagnostics using liquid biopsy samples. MethylBERT represents a significant advancement in read-level methylome analysis and enables accurate tumour purity estimation. The broad applicability of MethylBERT will enhance studies on both tumour and non-cancerous bulk methylomes. Mapping DNA methylomes in single cells is challenging, and thus studies using bulk samples remain common. Here, authors develop a transformer-based method for methylation pattern analysis to enhance bulk methylome deconvolution and cancer detection.