Statistical methods for differential peak detection in Hi-C data

摘要： HI-C是目前最流行的用于探测全基因组细胞内3D染色质组织的方法。 由于一维基因组距离较远的基因座在三维空间中往往紧密堆积在一起，因此增强子-启动子 相互作用可以发生在基因组的远端区域之间。重要的是，这种基因组结构保存得很好 跨细胞类型甚至物种，这种结构的失调被认为是ABER的来源之一。 RANT基因表达与阿尔茨海默氏症、自身免疫性疾病和癌症等疾病相关。因此， 有必要提供强大的方法来精确定位健康和健康之间的差异交互作用 以准确识别新的致病来源和潜在的治疗途径。 分析高空数据是具有挑战性的，因为数据中独特的空间结构，这意味着既有一维的 基因组距离依赖和3D空间依赖，需要仔细注意。这样做的统计工具 不考虑这些依赖项受到检测交互的能力降低的影响，尤其是 远端染色体区域。此外，用于在一对Hi-C数据集之间进行差分峰值检测的方法包括 不发达，可扩展到多个联合比较的方法完全缺失。我提议发表以下讲话 通过开发一种统计上严格的方法来检测Hi-C数据中的差值峰来解决这些问题 这两者都解释了Hi-C的标志性空间依赖结构，并扩展到多个联合比较 跨越生物条件(即细胞类型、细胞谱系或实验和对照类型)。我假设 这种方法将大大提高检测Hi-C样本中差异相互作用的能力。此外，通过软件 向公众开放，科学家将能够应用这些工具来识别发病的新驱动因素， 最终使fi有益于人类健康。我将在赞助商和合作伙伴的密切指导下开展这项工作 赞助商，分别拥有Hi-C数据的统计和生物学专业知识。

Abstract: Hi-C is currently the most popular assay used to probe 3D chromatin organization within the cell genome-wide. Because loci far away in 1D genomic distance are often packed close together in 3D space, enhancer-promoter interactions can occur between distal regions of the genome. Importantly, this genome structure is well-conserved across cell types and even species, and dysregulation of this structure has been implicated as a source of aber- rant gene expression associated with diseases such as Alzheimer's, autoimmune disorders, and cancer. Thus, it is necessary that powerful methods be made available to pinpoint differential interactions between healthy and diseased cells in order to accurately identify new sources of pathogenesis and potential pathways for treatment. Analysis of Hi-C data is challenging because the unique spatial structure in the data, which implies both a 1D genomic distance dependence and a 3D spatial dependence, requires careful attention. Statistical tools that do not account for these dependencies suffer from reduced power to detect interactions, especially those between distal chromosomal regions. Further, methods for differential peak detection between a pair of Hi-C datasets are underdeveloped, and methods that scale to multiple joint comparisons are wholly missing. I propose to address these problems by developing a statistically rigorous methodology for detecting differential peaks in Hi-C data that both accounts for Hi-C's hallmark spatial dependence structure and scales to multiple joint comparisons across biological conditions (i.e. cell types, cell lineages, or experimental and control types). I hypothesize that this approach will greatly boost power to detect differential interactions in Hi-C samples. Moreover, with software made available to the public, scientists will be able to apply these tools to identify new drivers of pathogenesis, ultimately benefitting human health. I will conduct this work under the close guidance of a sponsor and co- sponsor, respectively, with statistical and biological expertise with Hi-C data.