Protein structure determination from low-resolution experimental data

项目摘要 确定蛋白质的三维结构在生物学中至关重要， 为药物设计提供对生物机制和重要靶点的见解。虽然很高- 分辨率X射线衍射数据为许多人提供了细胞成分的原子视图 有趣的和生物相关的络合物，可能只能获得低分辨率 结构信息。低温电子显微镜和X射线结晶学，当应用时 对于大型、灵活的分子机器，通常会产生3-6？分辨率的数据。提取详细信息 来自这些数据的原子信息，对于理解功能、突变的影响或在 设计药物是不可能的，因为观察的数量很少，而观察的数量很大 构象空间蛋白可能采用。我建议开发计算方法，用于 从这些低分辨率的数据中提取高分辨率的原子模型，架起了 差距“与计算方法。 我提出的研究开发并扩展了我们实验室的自动推断方法 原子精度模型，来自这些“近原子”分辨率的实验数据来源。我们 在实验数据的指导下，开发新的构象采样方法来推断原子 在有相应的高分辨率数据的情况下以及在哪里都有信息 不。此外，我们还建议发展估计模型不确定性的方法；这些 对于理解从多大程度上可以从 特定的数据集。最后，为了进一步提高分辨率极限，我们开发了通用工具 生物分子力场优化。这些机器学习工具将允许开发 下一代力场，在扩展数据的分辨率限制方面至关重要，我们可以 推断原子细节。 拟议研究的总体目标是采用可靠和可获得的方法来确定 仅从稀疏的实验数据到原子精度的蛋白质结构。加在一起，这三个 这项提议的目的将大大提高我们推断原子相互作用的能力 从稀疏的实验数据。这将导致确定将揭示关键的结构 对生物医学上重要的蛋白质复合体如何发挥其功能以及如何发挥作用的见解 在人类疾病中是错误的。

Project Abstract Determination of a protein’s three-dimensional structure is of critical importance in biology, providing insights to biological mechanisms and important targets for drug design. While high- resolution X-ray diffraction data provides an atomic view of cellular components, for many interesting and biologically relevant complexes, it may only be possible to obtain low-resolution structural information. Both cryo-electron microscopy and X-ray crystallography, when applied to large, flexible molecular machines, often produce data of 3-6 Å resolution. Extracting detailed atomic information from this data, critical in understanding function, the effects of mutation, or in designing drugs is impossible due to the low number of observations and the large conformational space proteins may adopt. I propose to develop computational methods for extracting high-resolution atomic models from this low-resolution data, bridging the “resolution gap” with computational methods. My proposed research develops and extends our labs’ methods for automatically inferring atomic accuracy models, from these “near-atomic” resolution sources of experimental data. We develop novel conformational sampling methods, guided by experimental data, to infer atomic information both in cases where homologous high-resolution data is available, and where it is not. Additionally, we propose development of methods for estimating model uncertainty; these are critical in understanding to what degree structural conclusions may be made from a particular dataset. Finally, in pushing the resolution limit further, we develop general tools for biomolecular forcefield optimization. These machine-learning tools will allow development of a next-generation forcefield, critical in extending the resolution limit of data from which we can infer atomic details. The overall goal of the proposed research is robust and accessible methods to determine protein structures to atomic accuracy from only sparse experimental data. Combined, the three aims in this proposal will lead to dramatic improvements in our ability to infer atomic interactions from sparse experimental data. This will lead to determination of structures that will reveal key insights into how biomedically important protein complexes perform their function and what goes wrong in human disease.