ABI Innovation: Empirical Dynamics: A Next-Generation Approach For Uncovering Hidden Causal Links in Gene Expression

近15年前首次对33亿个碱基对的完整人类基因组进行了测序。尽管如此，基因科学距离理解这一详尽的“部分”如何组合在一起还有很长的路要走。该项目通过利用基因表达是一个时间和上下文相关的过程这一事实，预示着理解基因如何相互作用的基本数学方法发生了革命性的转变。例如，每个人都有产生褪黑素的遗传密码；然而，它的表达在一天中变化(取决于生物钟中的基因)，而且它对环境因素很敏感，比如光照。因此，表达的时间顺序和语境很重要。然而，目前理解表达可变性的方法依赖于基于相关性的非时态统计框架。他们假设，如果基因相互作用，它们要么总是正相关的(在所有样本中同时表达)，要么总是负相关的(只有当另一个没有表达时)，而与背景或不断变化的细胞环境无关。虽然这是一种方便的简化，但这种相关的方法显然是不完整的，可能会忽略一些基本过程，如阈值设置、制度转变和基因检查点，这些过程具体来自动态的、上下文相关的行为。这个项目将研究经验动态建模(EDM)作为一个新兴的框架，它明确地解释了表达事件的背景和时间序列。这项工作将是原理证明和应用于与乳腺癌相关的小鼠细胞分化途径的混合体。EDM的发展将是对现有方法的重要补充，能够显著提高生物信息学和系统生物学研究的效率，并揭示不相关的、因此当前方法不可见的重要基因调控中心。本项目将研究并进一步发展经验动态建模(EDM)，作为确定基因之间因果相互作用的概念框架，并产生可通过预测验证的机制理解。作为一种适应自然(非工程)非线性(上下文相关)互联系统的现实的方法，EDM非常适合利用最近才变得可行的时间显式基因组数据集。在第一阶段，该方法将用于构建酵母(S.cerevisiae)细胞周期中的基因相互作用网络，以探索一种显式的非线性和动态方法如何揭示当前分析方法所隐藏的信息。在此基础上，第二阶段将寻求开发和实施EDM方法(静态交叉映射)，以处理必然缺乏明确时间序列的单个单元数据。该方法将在基因电路的计算模型上开发和测试，然后应用于从小鼠干细胞分化序列中获得的单细胞数据。这项工作将通过与现有本体论的比较和由索尔克研究所的合作实验室进行的量身定制的实验来验证。软件将通过BioConductor：https://www.bioconductor.org/，分发所有其他材料(视频动画、互动演示等)。将在杉原实验室的网站上托管：http://deepeco.ucsd.edu.

The full human genome of 3.3 billion base pairs was first sequenced nearly 15 years ago. Nonetheless, genetic science is still a long way off from understanding how this exhaustive list of "parts" fits together. This project heralds a transformative shift in the basic mathematical approach to understanding how genes interact, by exploiting the fact that gene expression is a temporal and context dependent process. For example, every person has the genetic coding to produce melatonin; however, its expression varies through the day (keying on genes in the circadian clock) and it is sensitive to environmental factors like light exposure. Thus the temporal sequence and context of expression are important. However, current approaches to understanding variability in expression have relied on non-temporal statistical frameworks based on correlation. They assume that if genes interact, they will always either be positively correlated (simultaneously expressed among all samples) or negatively correlated (expressed only when the other is not) regardless of context or the changing cellular environment. Though a convenient simplification, such correlative approaches are obviously incomplete, and are likely to overlook essential processes such as thresholding, regime shifts, and gene check-pointing that specifically arise from dynamic, context-dependent behavior. This project will investigate empirical dynamic modeling (EDM) as an emerging framework that explicitly accounts for both the context and temporal sequence of expression events. The work will be a mixture of proof-of-principle and application to a cell differentiation pathway in mice associated with breast cancers. Development of EDM will provide an important complement to current approaches, with the capability to dramatically increase the efficacy of bioinformatics and systems biology research, and to reveal important regulatory hubs of genes that are non-correlated and thus invisible to current approaches.This project will investigate and further develop empirical dynamic modeling (EDM) as a conceptual framework to identify causal interactions among genes and produce mechanistic understanding that can be validated by prediction. As an approach that accommodates the reality of natural (non-engineered) nonlinear (context-dependent) interconnected systems, EDM is well-suited to leverage temporally-explicit genomic datasets that have only recently become feasible. In the first phase, the approach will be applied to construct gene interaction networks during the yeast (S. cerevisiae) cell cycle to explore how an explicitly nonlinear and dynamic approach can reveal information that is hidden to current analytical methods. Building on this, the second phase will seek to develop and implement an EDM approach (static cross-map) to single cell data that necessarily lack an explicit time sequence. The approach will be developed and tested on computational models of gene circuits, then applied to single cell data obtained from a stem-cell differentiation sequence in mice. Work will be validated by comparison to existing ontologies and by tailored experiments to be carried out by a collaborating lab at The Salk Institute. Software will be distributed through BioConductor: https://www.bioconductor.org/, all other materials (video animations, interactive demos, etc.) will be hosted on the Sugihara Lab website: http://deepeco.ucsd.edu.