SEI(BIO): Integration of Multimodal Experiments for Protein Structure

确定蛋白质的三维结构可以深入了解它们的进化起源、功能和机制。虽然通过传统方法确定原子细节结构可能昂贵、耗时，甚至不可行，但粗粒度结构表征通常足以提供重要的见解。快速近似蛋白质结构的多模态方法整合了来自许多来源的互补实验证据，以便验证和区分计算预测的结构。适当的实验，由于其速度、信息的多样性和不连贯的实验限制，包括交联，给出粗略的距离限制；诱变后的稳定性分析，在特定环境中表征残基的结构作用；溶液x射线散射，产生全局形状特性。多模态方法基于概率模型，该模型评估具有结构特征（距离、可访问性、整体形状）的数据的一致性。这种方法利用了可用方法的多样性和相对独立性，而不是寻求将单独的测量集成到一个总体物理模型中。然后，推理算法对结构和特征的后验分布进行推理，避免在单个答案中出现错误乐观，测量总体合理性，评估可用信息内容，并量化单个特征的不确定性。关联实验规划算法对给定的分析任务进行多模态实验优化（如选择交联子长度和特异性、确定最优突变位点），从而在最大化信息增益的同时有效利用实验资源。多模态整合机制正在开发、应用和测试来自不同方法的已发表数据，并被用于计划和解释针对未知结构的选定蛋白质的适当实验。研究生的积极参与扩大了教育活动和工具的可及性。

Determination of three-dimensional structures of proteins provides insight into their evolutionary origins, functions, and mechanisms. While determining atomic-detail structures by traditional methods can be expensive, time consuming, or even infeasible, coarser-grained structural characterization is often sufficient to provide significant insight. The multimodal approach to rapid, approximate protein structure integrates complementary experimental evidence from a number of sources in order to verify and discriminate among computationally predicted structures. Appropriate experiments, due to their speed, variety of information, and disjoint experimental limitations, include cross-linking, giving rough distance restraints; stability assays after mutagenesis, characterizing structural roles of residues in particular environments; and solution x-ray scattering, yielding global shape properties.The multimodal approach is grounded in probabilistic models that evaluate consistency of data with structural features (distances, accessibilities, overall shapes). This approach takes advantage of the diversity and relative independence of the available methods, rather than seeking to integrate separate measurements into an overarching physical model. Inference algorithms then reason about posterior distributions of structures and features, avoiding false optimism in a single answer, measuring overall plausibility, assessing the available information content, and quantitating uncertainty in individual features. Associated experiment planning algorithms optimize multimodal experiments (e.g. selecting cross-linker length and specificity, and identifying optimal mutation sites) for a given analysis task, so as to efficiently utilize experimental resources while maximizing information gain. The multimodal integration mechanism is being developed, applied, and tested with published data from individual methods, and is being used to plan and interpret appropriate experiments for selected proteins of unknown structure. Active participation of graduate students expands educational activities and tools will be accessible.