Data analysis tools for leveraging massive public data to improve hypothesis-driven research

项目摘要 科学成果的再现性和可复制性存在危机。这场危机是一个日益严重的根源， 在科学和大众媒体上都受到关注。这场危机如此严重，以至于美国国会目前正在 研究科学过程的可重复性。这场危机的核心是一系列问题，包括 样本量小，研究力度不足，数据分析师培训不足，无法直接利用先前的数据， 使用高通量技术对较小的假设驱动的实验进行统计分析。 技术的进步大大降低了收集高通量分子的成本和难度， 数据大量的原始数据越来越多地公开，但通常被纳入个人数据库。 NIGMS和其他研究人员在特定基础上进行分析。与此同时，运行一个设计好的， 假设驱动的研究并没有随着技术进步以同样的速度减少。它仍然是昂贵的， 识别、招募、收集和跟踪样本，即使高通量测量本身很便宜。 尽管现有的公开数据数量惊人，但进行统计推断仍然是常见的做法 在这些假设驱动的实验中，一项接一项的研究，只是间接地包括以前的数据，估计， 结果因此，这些研究的结果可能是高度可变的，不可靠的，或不可复制的。我们的团队专注于 开发统计方法、数据资源、软件和培训，使研究人员能够借鉴 从公共存储库、大规模数据生成项目和众包数据中获得经验， 在个体、假设驱动研究中改进推理。我们建议在我们的工作基础上， 统计数据来源，方法，软件和培训，促进和加快我们的生物和 医学合作者其结果将是一个研究社区，可以利用公共数据已经 为了提高功效，减少所需的样本量，并提高重复性， 许多新的假说驱动的发展和障碍的分子研究。

Project summary There is a crisis of reproducibility and replicability of scientific results. This crisis is an increasing source of concern both in the scientific and popular press. The crisis is so acute that the United States Congress is currently investigating reproducibility of the scientific process. At the heart of this crisis is a collection of problems including small-sample sizes, under-powered studies, under-trained data analysts and an inability to directly leverage prior results in the statistical analysis of smaller, hypothesis-driven experiments using high-throughput technologies. Advances in technology have dramatically reduced the cost and difficulty of collecting high-throughput molecular data. Large collections of raw data are increasingly publicly available but are usually incorporated into individual analyses by NIGMS and other investigators on an ad-hoc basis. Meanwhile, the other costs of running a designed, hypothesis-driven study have not decreased at the same speed with technological advances. It is still expensive to identify, recruit, collect, and follow up samples even if the high-throughput measurements themselves are cheap. Despite the incredible amount of available public data, it is still common practice to perform statistical inference in these hypothesis-driven experiments study-by-study, only indirectly including previous data, estimates, and results. So findings from these studies may be highly variable, unreliable, or unreplicable. Our group has focused on developing statistical methods, data resources, and software and training that allow researchers to borrow strength empirically from public repositories, large-scale data generation projects, and crowd-sourced data to improve inference in individual, hypothesis driven studies. We propose to build on our work in developing statistical data sources, methods, software and training that facilitate and speed the work of our biological and medical collaborators. The result will be a research community that can take advantage of public data already collected at a large cost to the NIH to improve power, reduce required sample sizes, and improve replication in many new hypothesis driven molecular studies of development and disorder.