Structure-based functional annotation of microbial genomes

摘要 现代生物学中最紧迫的挑战之一是翻译大量的 通过测序的最新进展已经获得的关于生物序列的信息 技术，转化为对生物系统行为的相应见解。确定功能和 蛋白质的生理作用仍然是这一挑战的主要组成部分;对于许多物种，特别是 非模式微生物，如微生物病原体，蛋白质组的一部分，包括注释不佳的 蛋白质可能接近50%，严重限制了我们的能力，甚至确定发病机制， 潜在的治疗目标。大量注释不佳的潜在生物学功能蛋白质， 重要性需要不断发展有效和可靠的计算方法， 蛋白质的功能注释。在过去的几年里，我们已经开发和应用了几种新的 全蛋白质组结构预测和细菌基因组功能注释的工作流程， 应用于实验室菌株E. coli K12和支原体JCVI-syn 3.0的最小基因组。我们 工作流程的特点是整合了结构信息（包括高精度蛋白质 结构预测），以及经典的方法，如序列同源性和 语法，以及最近的发展，例如包含基于深度学习的预测器;我们发现 总的来说，我们的工作流程提供了高度准确的功能注释，特别适用于“困难” 没有明确注释的同源物的蛋白质靶点。现在，我们将把重点转移到将我们的工具应用于 蛋白质组的细菌病原体，最初的重点是尿路致病性大肠杆菌。杆菌具体来说，我们将 继续发展我们的结构/功能预测能力，以进一步提高准确性， 提供丰富的信息（目标1），进行预测指导的生化表征可能 毒力基因，以评估预测性能和确定潜在的药物靶点（目的2），获得 蛋白质的实验结构被确定为困难的结构目标，这可能代表新的 折叠或不寻常的序列为已知的折叠（目的3），并测试可能的生理重要性 在体内小鼠模型中新鉴定的毒力因子（Aim 4）。收集的实验数据 目标2-4将继续与目标1下正在进行的方法开发相结合， 开发工具的性能和实用性。这一项目的成果将包括进一步改进 被广泛使用和引用的工具，用于快速结构/功能预测，鉴定特异性毒力 致病性E.大肠杆菌和初步的见解，他们如何可能针对 药物干预，以及潜在毒力因子的额外结构数据，这将有助于 基于结构药物设计和提高现有结构模板库的覆盖率，以指导未来 蛋白质结构与功能预测

Abstract One of the most pressing challenges in modern biology is that of translating the massive amounts of information on biological sequences that has been made available by recent advances in sequencing technologies, into corresponding insights into the behavior of biological systems. Determining the functions and physiological roles of proteins remains a major component of this challenge; for many species, especially non-model microbes such as microbial pathogens, the fraction of the proteome consisting of poorly annotated proteins may approach 50%, severely limiting our ability to even identify mechanisms of pathogenesis and potential therapeutic targets. The massive number of poorly annotated proteins of potential biological importance necessitates the ongoing development of efficient and reliable computational approaches for functional annotation of proteins. Over the past few years, we have developed and applied several new workflows for whole-proteome structure prediction and functional annotation of bacterial genomes, with applications to laboratory strain E. coli K12 and to the minimal genome mycoplasma JCVI-syn3.0. Our workflows are distinguished by the integration of structural information (including high-accuracy protein structure prediction) in functional annotations, alongside classical methods such as sequence homology and syntenty, and recent developments such as the inclusion of deep-learning based predictors; we find that collectively, our workflows provide highly accurate functional annotations that are especially useful for ‘difficult’ protein targets without clear annotated homologs. We will now shift our focus to applying our tools to the proteomes of bacterial pathogens, with an initial emphasis on uropathogenic E. coli. Specifically, we will continue to develop our structure/function prediction capabilities to further improve accuracy and increase the richness of information delivered (Aim 1), perform prediction-guided biochemical characterization of likely virulence genes to assess predictive performance and identify potential pharmaceutical targets (Aim 2), obtain experimental structures for proteins that are identified as difficult structural targets which likely represent novel folds or unusual sequences for known folds (Aim 3), and test the physiological importance of likely newly-identified virulence factors in an in vivo mouse model (Aim 4). The experimental data gathered under Aims 2-4 will be continuously integrated with the ongoing methods development under Aim 1 to maximize the performance and utility of the developed tools. The results of this project will include further improvements to widely used and cited tools for rapid structure/function prediction, identification of specific virulence determinants in uropathogenic E. coli and preliminary insights into how they may be targeted for pharmaceutical intervention, and additional structural data of potential virulence factors that will aid in structure-based drug design and improve coverage of existing structural template libraries to guide future protein structure and function prediction.