Causal inference of gene regulatory networks with application to breast cancer

描述（由申请人提供）：在基因调控及其对疾病影响的机制研究中，近年来主要分为两类研究。一方面，基因调控网络和蛋白质相互作用网络已经得到了广泛的研究，特别是在系统生物学中，遗传变异通常被忽视。另一方面，在全基因组关联研究中，已经发现了许多疾病的突变、插入和缺失以及拷贝数变异。因此，了解遗传变异如何通过基因调控网络影响疾病是非常有趣的。为了构建这些网络，至少有三个关键信息是重要的：基因表达、转录因子结合和基因型（特别是在表达数量性状位点上，即eqtl）。特别是后两者能够在网络构建中进行因果推理，尽管如何以概率和严格的方式使用它们尚未得到系统的探索。凭借我在贝叶斯统计方面的丰富经验，我的目标是开发统计模型和有效的计算策略，利用图形模型和因果推理的最新进展，构建涉及遗传变异和TF结合的因果调节网络。我将使用乳腺癌作为一种疾病模型，并将提出的方法应用于不同的亚型。推断的调节网络的拓扑特征可能提示乳腺癌亚型的潜在不同机制。通过提出的研究，我不仅将开发通用的分析方法来整合各种类型的高通量基因组学数据并提供开源软件，而且还将建立它们与疾病研究的相关性。凭借扎实的理论和应用统计学培训，以及与实验生物学家合作和处理各种生物数据的丰富经验，我的目标是实现这一转变

DESCRIPTION (provided by applicant): In the investigation of the mechanisms behind gene regulation and its impact on diseases, two lines of research have been largely separately carried out in recent years. On the one hand, gene regulatory networks and protein interaction networks have been under extensive study, especially in systems biology, where genetic variation is usually ignored. On the other hand, mutations, indels (insertions and deletions), and copy number variants have been identified for many diseases in genome-wide association studies. It is therefore of immense interest to understand how genetic variation influences disease through gene regulatory networks. To construct these networks, at least three key pieces of information are important: gene expression, transcription factor binding, and genotypes (especially at expression quantitative trait loci; that is, eQTLs). In particular, the later two enable causal inference in the network construction, although how to use them in a probabilistic and rigorous way has not been systematically explored. With my extensive experience in Bayesian statistics, I aim to develop statistical models and efficient computational strategies, drawing on recent advances in graphical models and causal inference, to construct causal regulatory networks involving genetic variation and TF binding. I will use breast cancer as a disease model and apply the proposed methodologies to different subtypes. Topological features of the inferred regulatory networks may suggest potentially different mechanisms in breast cancer subtypes. With the proposed research, I will not only develop general analysis methodologies to integrate various types of high-throughput genomics data and provide open-source software, but also establish their relevance to disease studies. With solid training in theoretical and applied statistics, as well as extensive experience collaborating with experimental biologists and working with a variety of biological data, I aim to make the transition from a statistician to a computational biologist and to become an independent investigator. I aspire to be not only an expert in developing sophisticated and rigorous statistical models and supplementing these models with efficient algorithms, but also a scientist capable of generating and testing my own hypotheses, either by myself or in collaboration with experimental biologists. The proposed K99/R00 award, involving one year of the mentored phase and three years of the independent phase, would greatly facilitate this transition, providing the unique opportunity for me to gain not only experience in genomic research in human, but also skills and experience in the wet lab, such that I can conduct some experiments on my own and eventually run an independent lab that focuses on computational research but also allows for experimental exploration.