Developing Computational Models on Single-Cell DNA-Methylation Data for Characterizing Functional Heterogeneity of Stem Cells in Mammalian Hematopoiesis

DNA甲基化是DNA中CpG二核苷酸可逆添加甲基的过程，是表观遗传调控的重要一层，对哺乳动物的细胞分化是不可或缺的。在造血干细胞分化的背景下，DNA甲基化的差异与造血干细胞（hsc）的长期命运偏差有关，而这种偏差在其他更动态的表观遗传层（包括转录组）上是不可见的。DNA甲基化的动态进一步与年龄相关的HSC功能下降以及白血病形成的早期步骤相关。然而，造血干细胞及其直系后代中dna甲基化模式的异质性以及相关的功能后果尚不清楚，主要是由于缺乏适当的单细胞方法。最近，分析单细胞DNA甲基化已经成为可能，但使用全基因组方法产生的数据是嘈杂和稀疏的。此外，需要结合谱系追踪方法来追踪造血干细胞的细胞命运，以便将这种dna甲基化动力学与其功能后果联系起来。在这个项目中，我将参与开发靶向单细胞dna甲基化测定（scTAM-seq）。这种有针对性的方法可以产生高达700个基因组位置的高分辨率剖面，测序工作量低，其特点是辍学率小，能够捕获谱系追踪条形码。scam -seq独有的挑战是潜在信息区域的选择。因此，我将开发可扩展的软件解决方案，使用我在博士期间共同开发的工具（例如，RnBeads, MeDeCom）来选择感兴趣的区域。为了进一步促进使用scam -seq对生物系统的分析，我将开发一个机器学习模型，用于去噪数据和提取单细胞甲基化状态。可以利用深度自动编码器等计算模型来理解由scam -seq生成的数据，并提取数据的低维可视化。使用这个工具集，我将探讨造血干细胞的血统偏差。特别是，我将研究小鼠模型系统中的DNA甲基化动力学，该系统允许跟踪造血干细胞的细胞命运和谱系潜力。总之，我将开发用于分析靶向单细胞dna甲基化数据的工具，并应用这些工具来表征造血过程中甲基化状态的功能异质性。

DNA methylation, the reversible addition of a methyl group to CpG dinucleotides in the DNA, is an important layer of epigenetic regulation and indispensable for cellular differentiation in mammals. In the context of the differentiation of blood stem cells, differences in DNA methylation have been associated with long-term fate biases of hematopoietic stem cells (HSCs) that are not visible on other, more dynamic epigenetic layers including the transcriptome. Dynamics of DNA methylation are further associated with the age-related decline in HSC function as well as early steps of leukemia formation. However, the heterogeneity of DNA-methylation patterns in HSCs and their immediate progeny as well as associated functional consequences are unknown, mostly due to the lack of appropriate single-cell methods. Recently, profiling DNA methylation of single cells has become feasible, but data generated using genome-wide approaches is noisy and sparse. Furthermore, an integration with lineage tracing approaches that track the cellular fate of HSCs is required for associating such DNA-methylation dynamics with its functional consequences. In this project, I will contribute to the development of a targeted single-cell DNA-methylation assay (scTAM-seq). This targeted approach allows for generating high-resolution profiles of up to 700 genomic positions with low sequencing effort, and is characterized by small dropout rates and the ability to capture lineage-tracing barcodes. A challenge exclusive to scTAM-seq is the selection of potentially informative regions. Thus, I will develop scalable software solutions for the selection of regions of interest using tools that I co-developed in my PhD (e.g., RnBeads, MeDeCom). To further facilitate the analysis of a biological system using scTAM-seq, I will develop a machine-learning model for denoising the data and for extracting single-cell methylation states. Computational models such as deep autoencoders can be leveraged for understanding the data generated by scTAM-seq and for extracting a low-dimensional visualization of the data. Using this toolset, I will explore lineage biases in HSCs. Particularly, I will investigate DNA methylation dynamics in a mouse model system that allows for tracking cell fates and lineage potential of HSCs. In summary, I will develop tools for the analysis of targeted single-cell DNA-methylation data and apply these tools to characterize the functional heterogeneity of methylation states in hematopoiesis.