Harnessing Computational and Structural Biology Platforms for Drug Discovery

药物发现是昂贵而费力的，尽管计算工具相对于实验工具更便宜，但在一个迭代和主观的过程中，它们代表了大量的时间成本。在2019冠状病毒病大流行期间，药物发现的计算方面尤为重要，最近的登月计划就是一个例子。最初是针对SARS-CoV-2的主要蛋白酶进行特别大的片段筛选，现在涉及众包一组易于合成和多样化的片段，以创建一种抗病毒药物化合物，这对长期管理大流行至关重要。登月计划由来自世界各地的科学家组成，他们总共提交了超过15,000个碎片。然后，汇集的计算机资源用于生成铅化合物，然后可以合成并测试其功效。这种基于片段的药物发现（FBDD）过程通常涉及使用筛选过程中的小分子命中（例如x射线晶体学筛选）并对其进行优化以产生线索。这些优化的片段可以显示出对目标的亲和力增加了多个数量级。FBDD允许合理和目标驱动的先导产生，该领域仍在发展，目前有数十种药物处于临床试验中。然而，到目前为止，FBDD所贡献的药物只有4种被推向市场。这个项目要回答的问题是：这种工作流程能否通过使用单个结构生物学/药物发现平台，以一种集成的方式，实现对特定细菌目标的自动化？自动化管道可以由非专业人员使用吗？将评估单个工具的可靠性、准确性和可扩展性，并使用python开发一种集成它们的方法。DeLinker是一个很有前途的（基于python的）工具的例子，它可以被合并到计算机辅助药物设计（CADD）管道中的自动化工作流程中。DeLinker是一种用于片段链接和支架跳跃的机器学习方法，它解决了缺乏3D生成片段链接软件的问题。它可以利用两个初始片段的空间信息产生链接，利用它们之间的距离和它们的相对方向产生初始训练库中没有的新链接。自动化的Python工作流程将通过优化两个细菌靶点的引线来验证，即单核细胞增生李斯特菌中的转录调节因子PrfA和结核分枝杆菌的毒素-抗毒素系统。单核细胞增生乳杆菌是一种食源性病原体，可引起李斯特菌病，每年都有重大疫情发生。一个潜在的目标是单核增生乳杆菌的毒力机制，并且是非杀菌的，因此相对于传统的抗菌方法具有更小的耐药机会。在单核增生乳杆菌中发现了一些已知的蛋白内“隧道”抑制剂，这些抑制剂使用环融合的2-吡啶酮杂环，通过结合和减弱PrfA来降低毒力，因此这个特定的靶标已经成熟，可以通过片段连接的方法来开发含有该片段的具有抗毒力特性的抑制剂。结核分枝杆菌（TB）是传染病致死的头号原因。毒素-抗毒素系统调节细胞过程，是治疗靶点，结核病中的毒素MbcT除非被抗毒素MbcA中和，否则是杀菌的，并导致细胞迅速死亡。寻找MbcTA（毒素-抗毒素）复合物的小分子抑制剂或灭活的MbcA抗毒素可能是对抗结核病的一条途径。

Drug discovery is expensive and laborious, and although computational tools are cheaper relative to their experimental counterparts, they represent a significant time cost in what is still an iterative and subjective process. Computational aspects of drug discovery are particularly relevant during the COVID-19 pandemic, exemplified by the recent Moonshot effort. This began as a particularly large fragment screening against SARS-CoV-2's main protease and now involves crowdsourcing a set of easily synthesised and diverse fragments in order to create an anti-viral drug compound that will be vital for long term management of the pandemic. The Moonshot effort is composed of scientists from all over the world who have submitted a combined total of over 15,000 fragments. Pooled computer resources are then used to generate lead compounds, which can then be synthesised and tested for efficacy. This process of fragment-based drug discovery (FBDD), in general, involves using small molecule hits from a screening process (e.g. X-ray crystolographic screening) and optimising them to produce leads. These optimised fragments can show an increased affinity for their targets by multiple orders of magnitude. FBDD allows rational and target-driven lead generation, and the field is still growing with dozens of drugs currently in clinical trials. However as of yet only 4 drugs that FBDD has contributed to, have been brought to market. The question this project sets out to answer is:Can this sort of workflow be automated by using individual structural biology/drug discovery platforms in an integrated way for particular bacterial targets? Can the automated pipeline be used by a non-specialist? Individual tools will be assessed for reliability, accuracy and extensibility and a method of integrating them will be developed, using python. An example of a promising (Python-based) tool that can be incorporated into an automated workflow in the Computer-Aided Drug Design (CADD) pipeline is DeLinker. DeLinker is a machine learning approach for fragment linking and scaffold hopping, which addresses the lack of 3D generative fragment-linking software. It can produce linkers using spatial information of two initial fragments, utilising the distance between them and their relative orientations to produce novel linkers not in the initial training database.The automated Python workflow will be validated by optimising leads for two bacterial targets, the transcriptional regulator PrfA in Listeria monocytogenes and the toxin-antitoxin system of Mycobacterium tuberculosis. L. monocytogenes is a food-borne pathogen that causes listerosis, with major outbreaks occuring on an annual basis. A potential target is the virulence machinery in L. monocytogenes and is non-bactericidal, so has less chance of resistance relative to traditional antibacterial approaches. There are known inhibitors for an intraprotein 'tunnel' previously identified in L. monocytogenes using ring-fused 2-pyridone hetero-cycles that reduced virulence by binding and attenuating PrfA, so this particular target is ripe for a fragment-linking approach to develop inhibitors with antivirulence properties, containing that moiety. M. tuberculosis (TB) is the number one cause of death from infectious disease. Toxin-antitoxin systems regulate cellular processes and are therapeutic targets, and the toxin in TB, MbcT, is bactericidal unless neutralized by its antitoxin MbcA, and causes rapid cell death. The search for a small molecule inhibitor for the MbcTA (toxin-antitoxin) complex or the inactivate MbcA antitoxin could be an avenue to combat TB.