Statistical computation for deciphering biological outcomes in proteins and cells

我们正在目睹实验技术的进步，这些技术使快速收集蛋白质和细胞水平的生物数据成为可能。已知3-D蛋白质结构的蛋白质数据库一直在快速增长。改进后的成像技术可以更详细地跟踪细胞运动。许多动力系统的组成部分现在可以更精确和更接近实时地测量。因此，我们可以访问的数据集不仅很大，而且具有很高的复杂性和结构。这些科学发展反过来又为利用这些数据来回答重要生物学问题的新方法的发展提供了肥沃的土壤。统计学家在量化科学过程中出现的不确定性方面发挥着关键作用，他们建立模型和方法，产生新的见解，提高我们预测生物学结果的能力。本提案推进了一项研究计划，旨在发展必要的分析和计算方法，以促进我们对控制蛋白质、基因和细胞的复杂生物机制以及它们之间相互作用的科学理解。根据我们目前掌握的数据，需要解决的一些具体问题如下。如何将统计学中的抽样方法与数据驱动学习相结合，进一步提高我们从氨基酸序列预测三维蛋白质结构的能力？我们如何才能更好地理解细胞分子机制与细胞水平结果（如运动性和迁移）之间的联系？我们如何才能更好地将模型拟合到从监测生物系统中收集到的嘈杂数据中？总之，我们期望整合这些相互关联的生物学方面的知识-蛋白质，基因和细胞-并为他们的研究提供坚实的统计基础。待开发的方法将结合现代统计计算技术，特别是贝叶斯推理和蒙特卡罗方法。我们还将充分利用现代硬件进行计算，利用大规模并行化以及CPU和GPU计算。特别是，处理这些生物“大数据”需要高效且精心定制的算法，这些算法还要符合时间和计算的实际限制。拟议研究的一个重要的长期目标是其实际的生物医学影响。在药物发现中，人们对设计氨基酸序列以实现特定的三维结构和功能越来越感兴趣。在病理学上，异常的迁移信号可能导致错误的细胞迁移；因此，彻底了解细胞迁移机制在医学和治疗领域，包括癌症研究是必不可少的。

We are witnessing advances in experimental techniques that have enabled the rapid collection of biological data at the levels of proteins and cells. The Protein Data Bank of known 3-D protein structures has been growing quickly. Improved imaging techniques permit cellular motions to be tracked with greater detail. The components of many dynamical systems can now be measured more precisely and closer to real-time. As a result, we have access to datasets that are not only large, but with a high level of complexity and structure. These scientific developments, in turn, provide fertile ground for the development of novel approaches that leverage these data to answer important biological questions.  Statisticians have a key role in quantifying uncertainty that arises in the scientific process, by building models and methods that yield new insights and improve our capacity to make predictions of biological outcomes. This proposal advances a research program for developing the necessary analytical and computational methods to advance our scientific understanding of the complex biological mechanisms governing proteins, genes, and cells, and the interplay between them. Some specific questions to be addressed, in light of data currently available to us, are as follows.  How can sampling methods in statistics be adapted and combined with data-driven learning to further improve our ability to predict 3-D protein structure from amino acid sequence?  How can we better understand the links between molecular mechanisms in cells with cellular-level outcomes such as motility and migration?  How can we better fit models to noisy data collected from monitoring biological systems?  Together, we anticipate integrating knowledge from these interconnected biological facets - proteins, genes, and cells - and providing solid statistical foundations for their study. The methods to be developed will incorporate techniques of modern statistical computation, especially from the realms of Bayesian inference and Monte Carlo methods.  We will also fully utilize modern hardware for computation, leveraging mass parallelization as well as both CPU and GPU computing.  In particular, handling these biological "big data" requires efficient and carefully-tailored algorithms that also fit within practical constraints of time and computing. An important long-term goal of the proposed research is its practical biomedical impacts. In drug discovery, there is increasing interest in designing amino acid sequences to achieve specific 3-D structure and function.  In pathology, abnormal migratory signals may cause erroneous cell migrations; a thorough understanding of cell migration mechanisms is therefore essential in medical and therapeutic domains, including cancer research.