Reliable model building for cryo-EM

项目摘要 由实验结构生物学产生的原子模型被用来理解细胞过程 分子水平，也用于合理设计药物和治疗方法。低温电子显微镜单件 粒子重建(Cryo-EM SPR)基于对图的解释来生成这样的原子模型 库仑势是通过对数千个弱图像求平均而获得的，每个图像包含 在薄薄的冰层中悬浮的单个大分子的投影快照。因此，低温电磁地图 表示结构状态的平均值，通常具有有限且不均衡的质量 低温电磁SPR中使用的颗粒的灵活性及其在冰中的非随机分布称为首选 定位。在由于缺乏细节(分辨率)而导致质量较低的地图中构建和重建结构模型 或者，采样和重建问题是一个挑战。 在这项提案中，我们将通过开发和实施自动化和 全面的模型构建和验证，以及精心设计的实验数据反馈回路 分析。我们的模型构建方法生成的初始和中间模型将有助于低温电磁SPR 通过在重建的所有步骤稳定计算的收敛，但不引入偏差， 其中明确地解决了偏差消除和量化问题。目标1将重点介绍将使用从头算的方法 改进低温电磁SPR的预测模型或实验模型，以便无需引入即可使用 偏见。在目标2中，用于改进和验证结构模型的新的信息权重将是 为解决低温电磁SPR中信息质量参差不齐的问题而开发和实施。目标3将重点放在 开发和实现了直接将二维分类与模型建立相结合的方法。最后，目标4将 目标分析和内部运动建模，以指导构造解释和改进 重建的解决方案。结果将被纳入商业分布的套装CryoEMMA中。 这些方法的竞争优势来自于它们能够提供高度信息量的3D 存在严重择优取向和单粒子信噪比低于 在目前的方法中。

Project Abstract Atomic models generated by experimental structural biology are used to understand cellular processes at the molecular level, and also for rational design of drugs and treatments. Cryogenic electron microscopy single particle reconstruction (cryo-EM SPR) generates such atomic models based on interpretation of maps of Coulomb potential that are obtained by averaging many thousands of weak images, with each image containing projection snapshots of individual macromolecules suspended in a thin layer of ice. Consequently, cryo-EM maps represent averages of structural states and frequently have limited and uneven quality resulting from both flexibility of particles used in cryo-EM SPR as well as their non-random distributions in ice called preferred orientation. Building and rebuilding structural models in maps that are low quality due to lack of detail (resolution) or having sampling and reconstruction problems is a challenge. In this proposal, we will address these challenges by developing and implementing methods for automatic and comprehensive model building and validation with carefully designed feedback loops to experimental data analysis. The initial and intermediate models generated by our model building methods will aid cryo-EM SPR projects by stabilizing convergence of computations at all steps of reconstruction, but without introducing bias, with bias removal and quantification being addressed explicitly. Aim 1 will focus on methods that will use ab initio predicted models or experimental models to improve cryo-EM SPR so that they can be used without introducing bias. In Aim 2, a new weighting of information for refinement and validation of the structural models will be developed and implemented to account for uneven quality of information in cryo-EM SPR. Aim 3 will focus on developing and implementing methods for directly coupling 2D classification to model building. Finally, Aim 4 will target analysis and modeling of internal motions to guide the structural interpretation and to improve the resolution of reconstructions. The results will be incorporated into the commercially distributed suite cryoEMMA. The competitive advantage of these approaches arises from their ability to provide highly informative 3D reconstructions in the presence of severe preferred orientation and for single particle signal-to-noise lower than in the current approaches.