Liu Yang, Taha Hash Brown, Zhang Qiuyang, Pan Ziwei, Chatzipantsiou Christina, Wade Emma, Slocum Thatcher, Karuturi Lasya, Zhao Yue, Karmakar Shilpita, Li Sheng
Department of Cancer Biology, Keck School of Medicine, Norris Comprehensive Cancer Center, University of Southern California, Los Angeles, CA, USA.
The Jackson Laboratory for Genomic Medicine, Farmington, CT, USA.
bioRxiv. 2025 Jul 4:2025.06.29.662079. doi: 10.1101/2025.06.29.662079.
Nanopore long-read sequencing has expanded the capacity of long-range, single-base, and single-molecule DNA-methylation (DNAme) detection and haplotype-aware allele-specific epigenetic phasing. Previously, we benchmarked and ranked the robustness of seven computational tools for DNAme detection using nanopore sequencing. The top performers were Megalodon, Nanopolish, DeepSignal and Guppy. However, these algorithms exhibit lower performance at regions with discordant non-singleton DNAme patterns compared to genome-wide regions. Furthermore, long-read sequencing analysis of mammalian genomes requires higher computational resources than next-generation sequencing. To address these issues, we developed a NANOpore Methylation (NANOME) a consensus DNAme predictive model using XGBoost, which integrates the output of Megalodon, Nanopolish, and Deepsignal for analyzing data obtained using Oxford Nanopore Technologies (ONT). NANOME enhanced DNAme detection precision (mean square error) at single-base resolution by 11% and improved accuracy (F1-score) at single-molecule resolution by 2.4% for human B-lymphocyte European cell lines (NA12878). The consensus model also detected ~200,000 more CpGs than all three tools. Combing variant calling and long-read phasing, NANOME can detect haplotype-aware allele-specific DNAme in known imprinting controls in resolved and previously unresolved regions. We conducted haplotype-aware methylation detection on the T2T genome for dataset NA12878, revealing significant variations in differentially methylated region (DMR) density between gap and non-gap regions. Overall, NANOME represents a significant step forward in DNAme detection and long-range epigenetic phasing, offering a robust and accessible tool for researchers studying the epigenome.
纳米孔长读长测序扩展了长距离、单碱基和单分子DNA甲基化（DNAme）检测以及单倍型特异性等位基因表观遗传定相的能力。此前，我们对使用纳米孔测序进行DNAme检测的七种计算工具的稳健性进行了基准测试和排名。表现最佳的是Megalodon、Nanopolish、DeepSignal和Guppy。然而，与全基因组区域相比，这些算法在具有不一致非单例DNAme模式的区域表现较低。此外，哺乳动物基因组的长读长测序分析比下一代测序需要更高的计算资源。为了解决这些问题，我们开发了一种纳米孔甲基化（NANOME）共识DNAme预测模型，该模型使用XGBoost，整合了Megalodon、Nanopolish和Deepsignal的输出，用于分析使用牛津纳米孔技术（ONT）获得的数据。对于人类B淋巴细胞欧洲细胞系（NA12878），NANOME在单碱基分辨率下将DNAme检测精度（均方误差）提高了11%，在单分子分辨率下将准确率（F1分数）提高了2.4%。该共识模型还比所有三种工具多检测到约200,000个CpG。结合变异检测和长读长定相，NANOME可以在已解析和先前未解析区域的已知印记控制中检测单倍型特异性等位基因DNAme。我们对数据集NA12878的T2T基因组进行了单倍型特异性甲基化检测，揭示了间隙区域和非间隙区域之间差异甲基化区域（DMR）密度的显著差异。总体而言，NANOME在DNAme检测和长距离表观遗传定相方面向前迈出了重要一步，为研究表观基因组学的研究人员提供了一个强大且易于使用的工具。
