Removing statistical bottle-necks in data analysis for the ENCODE Consortium

描述（由申请人提供）：ENCODE联盟产生的数据构成了一个前所未有的机会，可以对人类基因组的功能和结构进行生物医学推断。在近1000个全基因组分析中，每个碱基对的公开信息深度是惊人的，在下一轮的合作中，这种深度可能会呈几何级数增长。在本提案中，我们描述了统计方面的挑战，如果这些挑战得到满足，将大大提高分析工作组（AWG）从ENCODE数据中进行功能生物学推断的能力。我们将解决这些挑战，作为具有内置实验验证功能的统计“研究和开发”组件，并将为AWG提供我们开发的统计工具的有用软件实现，以及基于我们的输入网络的实验验证的这些工具的迭代改进。具体而言，我们将：1)开发降维方法，帮助AWG进行数据可视化、总结和预测研究（例如，从染色质数据预测转录）；2)开发新的ENCODE分析的复杂生物系统定量网络模型；3)进行有针对性的生物验证试验，旨在询问模型中重要的低维结构，并反馈以改进模型结构和性能。我们的降维方法将帮助生物学家从高维基因组学数据中解释和制定假设，我们的网络模型将有助于构建可解释的预测算法，直接导致可测试和可量化的假设，我们的验证分析将确保从我们的工具中得出的推论提供有意义的生物学见解。正如我们作为ENCODE和modENCODE数据分析中心的一部分所做的那样，我们将与AWG密切合作，以确保我们的软件实现对联盟立即和最大限度地有用，并且我们在网络推理和降维方面的整个工作过程都集中在对联盟利益至关重要的生物学问题上。

DESCRIPTION (provided by applicant): The data generated by the ENCODE Consortium constitutes an unprecedented opportunity to make biomedical inferences about the function and structure of the human genome. With nearly 1000 genome-wide assays, the depth of information now publicly available about each base-pair is staggering, and in the next round of consortium work this depth is likely to increase geometrically. In this proposal, we describe statistical challenges that, if met, will substantially enhance the capacity of the Analysis Workin Group (AWG) to make functional biological inferences from ENCODE data. We will tackle these challenges, serving as a statistical "Research and Development" component with built-in experimental validation capabilities, and will provide the AWG with useful software implementations of the statistical tools we develop, as well with iterative refinements of these tools grounded in experimental validations of our imputed networks. In particular, we will: 1) develop methods of dimension reduction that will aid the AWG in data visualization, summarization, and prediction studies (e.g. the prediction of transcription from chromatin data); 2) develop new quantitative network models of complex biological systems assayed by ENCODE; and 3) conduct targeted biological validation assays designed to interrogate important low-dimensional structures in our models and to feed back to improve both model structure and performance. Our approaches to dimension reduction will aid biologists in interpreting and formulating hypotheses from high-dimensional genomics data, our network models will facilitate the construction of interpretable predictive algorithms that lead directly t testable and quantifiable hypotheses, and our validation assays will ensure that inferences derived from our tools provide meaningful biological insights. As we did as part of the ENCODE and modENCODE data analysis centers, we will work closely with the AWG to ensure that our software implementations are immediately and maximally useful to the consortium, and that the overall course of our work on network inference and dimension reduction is focused around biological questions central to the interests of the consortium.