Novel bioinformatics methods to detect DNA and RNA modifications using Nanopore long-read sequencing

项目总结 DNA修饰，如DNA上的5-甲基胞嘧啶(5mC)和5-羟甲基胞嘧啶(5hmC)，以及 RNA修饰，如mRNA上的N6-甲基腺苷(M6A)，已被认为与基因调控和 人类疾病。人工合成的碱基类似物，如BrdU、Edu和IDu，已被用作基因组标记 研究基本的生物过程。然而，检测DNA/RNA修饰的传统方法 依靠间接读数(如亚硫酸氢盐治疗或免疫沉淀)，不能检测重复区域(由于 到使用短读)，并遭受各种技术偏见。同时对DNA/RNA进行直接测序 牛津纳米孔平台可以解决这些技术限制，迫切需要开发可靠的 从离子流数据中检测常见DNA/RNA甲基化的生物信息学方法，具有以下能力 延伸到罕见的修改形式。我们多年来一直致力于计算工具的开发 用于长时间读取测序数据的信号电平分析。我们开发了能够综合检测的NanoMod 将DNA修饰引入复制细胞，以及DeepMod，它使用深度神经网络来预测 5mC直接来自纳米孔测序的离子电流信号。在目前的建议中，我们将：(1)发展 LongReadSum，它将由多线程C++实现，带有各种格式的模块(Fasta， FASTQ、FAST5、BAM、POD5等)，用于超快质量控制(QC)和来自纳米孔的信号汇总 测序。信号汇总过程生成用户指定的特征向量 用于调用修改的其他下游机器学习工具。(2)开发modDNA，我们将在那里使用 连接主义时态分类(CTC)和转换器，两种神经网络模型，称为修正 5 mC、5 hmC和6 mA等底座。此外，我们将调整计算流水线以减少 表示甲基化测序(RRMS)数据，使在一个单一的分析人类基因组 Minion Flowcell。(3)开发了一种结合了先前基因组特征和 从头开始和基于模型的上下文相关特征(例如，基因3‘端的丰富) 检测RNA m6A修饰和其他罕见的修饰。(4)验证和改进计算 通过对数据集进行基准测试的工具。我们将对癌症样本进行纳米孔DNA测序 来自临床诊断实验室的配对甲基化图谱，以及含有或不含有5-羟色胺的小鼠参考细胞系 氮杂-2‘-脱氧胞苷(甲基化抑制剂)治疗。我们将在参考文献上进行mRNA直接测序 具有或不具有METTL3/METTL14基因敲除，或具有体外转录，或具有或不具有KSHV的细胞株 感染会改变表位转录图谱。拟议项目的成功完成带来了 通过纳米孔测序检测DNA/RNA修饰的计算工具箱，提供参考 数据集，并极大地促进了我们对人类表观基因组和 墓志铭。

PROJECT SUMMARY DNA modifications, such as 5-methylcytosine (5mC) and 5-hydroxymethylcytosine (5hmC) on DNA, as well as RNA modifications, such as N6-methyladenosine (m6A) on mRNA, have been implicated in gene regulation and human diseases. Synthetic base analogs, such as BrdU, EdU and IdU, have been used as genomic markers to study fundamental biological processes. However, conventional approaches to detect DNA/RNA modifications rely on indirect readout (such bisulfite treatment or immunoprecipitation), cannot assay repetitive regions (due to the use of short reads), and suffer from various technical biases. While direct DNA/RNA sequencing on the Oxford Nanopore platform can address these technical limitations, there is an urgent need to develop reliable bioinformatics methods to detect common DNA/RNA methylations from ionic current data, with the ability to extend to rare forms of modifications. We have years of dedication to the development of computational tools for signal-level analysis of long-read sequencing data. We developed NanoMod which detects synthetically introduced DNA modifications into replicating cells, and DeepMod which uses a deep neural network to predict 5mC directly from ionic current signals from Nanopore sequencing. In the current proposal, we will: (1) Develop LongReadSum, which will be implemented by multi-threaded C++ with modules for diverse formats (FASTA, FASTQ, FAST5, BAM, POD5, etc), for ultrafast quality control (QC) and signal summarization from Nanopore sequencing. The signal summarization procedure generates user-specified feature vectors that can be used by other downstream machine-learning tools for calling modifications. (2) Develop ModDNA, where we will use connectionist temporal classification (CTC) and transformers, two neural network models, to call modified bases such as 5mC, 5hmC and 6mA. Additionally, we will adapt the computational pipeline to reduced representation methylation sequencing (RRMS) data, which enables assaying a human genome in one single MinION flowcell. (3) Develop ModRNA, an integrative model which combines prior genomic features with context-dependent features (for example, enrichment in 3’ end of genes) for both de novo and model-based detection of RNA m6A modifications and other rare modifications. (4) Validate and improve the computational tools via benchmarking data sets. We will perform Nanopore DNA sequencing from cancer samples with paired methylation profiles from clinical diagnostic labs, as well as mouse reference cell lines with or without 5- Aza-2’-deoxycytidine (methylation inhibitor) treatment. We will perform direct mRNA sequencing on reference cell lines with or without METTL3/METTL14 knockdown, or with in vitro transcription, or with and without KSHV infection which alters epitranscriptomic profiles. Successful completion of the proposed project delivers a computational toolbox for DNA/RNA modification detection via Nanopore sequencing, provide reference datasets to the community, and greatly facilitates our understanding of the human epigenome and epitranscriptome.