Scientific Findings across Multiple Environments: Replication, Robustness, and Equity in Genetic Association Studies

该项目的动机是需要探索在多种环境/人口中感兴趣的科学问题，以及这项任务在方法和成果传播方面提出的挑战。一方面，可复制性是科学的基石，任何发现都不能被认为是既定知识库的一部分，除非导致它的实验/观察至少在略有不同的情况下重复。另一方面，环境的过度实质性变化可能会影响最初发现背后的机制，因此，结果的变化不会使其无效，而只是强调了解释异质性的必要性。确定何时积累的有利于发现的证据足以认为它是确凿的，并确定一种模式在不同环境中的可变性是具有挑战性的任务。这一研究项目旨在推进用于解决这些问题的方法和用于向更大的科学家社区和公众传播研究结果的工具。该项目包括外联和交流活动，以及通过参与研究对研究生进行培训。研究人员和合作者的目标是开发新的统计方法，以衡量一项发现在多项研究中的可重复性，并评估与其在不同环境中相关的不确定性。他们将利用最新的方法论发展，例如e值的概念、保角预测区间和仿冒推理。利用研究结果的稳健性和因果机制之间已经建立的联系，该项目旨在为用户提供感兴趣的结果的预测模型，这些模型是可解释的，已在可变条件下得到验证，并具有可观的样本外性能。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

This project is motivated by the need to explore scientific questions of interest in multiple environments/populations and the challenges that this task presents both in terms of methods and communication of results. On the one hand, replicability is the cornerstone of science, and no finding can be considered part of the established corpus of knowledge unless the experiment/observation that led to it has been repeated under at least slightly different circumstances. On the other hand, overly substantial changes in environment might influence the very mechanism behind the original finding, so that, rather than invalidating it, variation of a result simply underscores the need to account for heterogeneity. Determining when the evidence accumulated in favor of a finding is enough to consider it corroborated and determining the variability of a pattern across different environments are challenging tasks. This research project aims to advance both the methods used to address these questions and the tools employed to communicate findings to the larger community of scientists and the public. The project includes outreach and communication activities, as well as training of graduate students through their involvement in the research. The investigator and collaborators aim to develop new statistical methods to measure the replicability of a finding across multiple studies and to evaluate the uncertainty associated with it in different environments. They will leverage recent methodological developments, such as the notion of e-values, conformal prediction intervals, and knockoffs inference. Exploiting the links that have been established between robustness of findings and causal mechanisms, this project intends to provide users with predictive models for outcomes of interest that are interpretable, have been validated across variable conditions, and have appreciable out-of-sample performance.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.