Causal Effect Estimation of Regulatory Molecules

项目摘要/摘要 转录因子和microRNAs是控制信使RNAs的基本调节分子 (信使核糖核酸)，并已知在人类疾病中调节失调。每个RM可能会影响多条途径 这既是一种祝福，也是一种诅咒。如果一种疗法针对适当的rm，它可以从 多条战线，提高药效。另一方面，靶向治疗可能会导致严重的不良反应。 由于我们对RM操纵的下游因果影响的了解有限。尽管本地绑定 从理论上和实验上研究了单个均方根与它们的靶之间的相互作用 这种结合对转录组的功能影响尚不清楚。在这里，我提出了统计机 预测观察到的基因表达变异性的学习技术和因果推理方法 只使用调节分子，并估计它们对整个 转录组。为了实现这一目标，我在目标1中首先构建了一个多响应预测模型来预测 整个转录组只使用RMS。这一目标具有挑战性，因为响应向量的维度 超过了样本的数量，我将使用高维统计中的技术来解决这个问题 问题。在目标2中，我将超越预测性建模，通过估计RMS对转录组的因果影响 使用不变的因果预测。我将利用快速增长的文献，将因果推理联系在一起 跨不同数据源的不变预测精度，以推断RMS对mRNA的因果影响。 在目标3中，开发了RMS对基因调控贡献的预测性和因果模型 R00阶段，我将重点介绍双/无偏机器学习的最新进展，它允许 使用可扩展的机器学习方法可靠地估计RMS对转录的因果效应。我的 拟议的研究将把最先进的统计机器学习和因果推理技术带到 通过提供对全球角色的洞察，进行基因组学研究并帮助设计更有效的靶向治疗 RMS在基因表达调控中的作用。在该奖项的培训阶段，在我杰出的 指导团队和科学咨询委员会，我将获得分子生物学和基因组学方面的专业知识，同时 完善我的因果推理和机器学习知识。俄亥俄州立大学综合 癌症中心-詹姆斯医院和数学生物科学研究所将为我提供理想的 跨学科的环境，架起数据科学和基因组学的桥梁，并将帮助我实现我的职业生涯 发展目标和过渡到终身教职的教师职位。

Project Summary/Abstract Transcription factors and microRNAs are essential regulatory molecules (RM) that control messenger RNAs (mRNA) and are known to be dysregulated in human diseases. Each RM may affect multiple pathways of the cell which is both a blessing and a curse. If a therapy targets the proper RM, it can attack the disease from multiple fronts and increase efficacy. On the other hand, targeted therapy may result in serious adverse effects due to our limited knowledge of the downstream causal effect of RM manipulation. Although the local bindings between single RMs and their targets have been studied computationally and experimentally, the intensity of functional consequences of such bindings on the transcriptome is unclear. Here, I propose statistical machine learning techniques and causal inference methods to predict the observed variability of gene expression using only regulatory molecules and estimate their downstream causal effect on the entire transcriptome. To achieve this goal, I start in Aim 1 by building a multi-response predictive model to predict the whole transcriptome using only RMs. This goal is challenging because the dimension of the response vector is more than the number of samples and I will use techniques from high-dimensional statistics to address this issue. In Aim 2, I will go beyond predictive modeling by estimating the causal effect of RMs on the transcriptome using invariant causal prediction. I will leverage the rapidly growing literature which connects causal inference to invariant prediction accuracy across heterogeneous data sources to infer the causal effect of RMs on mRNA. Having developed both predictive and causal models of RMs contribution to gene regulation, in Aim 3 during the R00 phase, I will focus on the most recent advances in double/debiased machine learning which allows the use of scalable machine learning methods for reliable estimation of causal effect of RMs on transcription. My proposed research will bring the most advanced statistical machine learning and causal inference techniques to genomics research and help design more effective targeted therapies by providing insights into the global role of RMs in gene expression regulation. During the training phase of the award, with the support of my outstanding mentoring team and scientific advisory committee, I will gain expertise in molecular biology and genomics while perfecting my knowledge of causal inference and machine learning. The Ohio State University Comprehensive Cancer Center – James Hospital and the Mathematical Biosciences Institute will provide me with the ideal interdisciplinary environment to bridge data science and genomics and will help me achieve my career development goals and transition to a tenure-track faculty position.