Data-driven approaches for fragment merging

该项目的重点是基于片段的药物发现（FBDD），涉及针对感兴趣的目标筛选低分子量化合物，以找到可以优化成为铅样分子的化学起点。拟议的工作位于FBDD管道的领先优化阶段，该阶段着眼于如何利用晶体学片段命中及其产生的结构信息来提出更大的药物样化合物，这些化合物以更高的亲和力与靶标结合。片段细化有几种方法，包括片段生长、链接和合并。与其对应物相比，片段合并是一种相对未开发的技术，但具有找到有效分子的潜力。现有的合并设计策略包括药物化学家的手动设计，这是缓慢的，不能扩展到大型数据集，和从头设计，这往往会导致缺乏合成可及性的分子，因此是困难和昂贵的追求。因此，该项目的目的是使用基于知识的方法来提出合成可行的片段合并，允许从片段命中到铅样化合物的快速和廉价的进展。将探索各种数据驱动的方法，包括数据库开发和基于AI的技术，这些技术可以协同使用来提出新的化合物。前者的使用已在轮换项目期间得到证明，该项目涉及使用片段网络，这是一个包含目录化合物的图形数据库，以创建一个能够找到和过滤片段合并的管道，可以优先进行进一步筛选。已经确定了通过提高该工具的效率和增加所发现分子的多样性来扩展这项工作的几种途径。本项目提出的化合物也将有机会进行实验验证。XChem目前正在探索机器人平台上的自动合成规划和执行，这项工作的一个关键组成部分将是确定可以在现有合成库中制备的化合物。整合整个管道有可能对提高我们在药物开发中开发有效配体的速度产生重大影响，从而减少设计-制造-测试周期所需的迭代次数。拟议的工作（将与另一个哲学博士项目一起存在，重点是管道的机器人方面）将涉及与Vernalis和LifeArc的工业合作;确切的工业主管有待确认。该项目福尔斯EPSRC的“计算和理论化学”研究领域，与该理事会在该领域的战略有着密切的联系。这项研究是高度跨学科的，将涉及与光束线科学家，药物化学家和自动化专家的合作。如上所述，在该项目期间开发的软件将对药物发现社区产生直接的、可操作的影响，产生可以购买或容易合成的分子，以进行进一步筛选。提供该项目的源代码和数据将允许其他人将这种技术应用于新的目标，并与他们自己开发的算法进行比较。

This project focuses on fragment-based drug discovery (FBDD), involving the screening of low-molecular-weight compounds against a target of interest to find chemical starting points that can be optimized to become lead-like molecules. The proposed work is situated within the lead optimization stage of the FBDD pipeline, which looks at how crystallographic fragment hits and the structural information they yield can be exploited to propose larger, drug-like compounds that bind to a target with increased affinity. There are several approaches to fragment elaboration, including fragment growing, linking and merging. Fragment merging is a relatively unexplored technique compared with its counterparts yet has the potential to find potent molecules. Existing strategies for merge design include manual design by a medicinal chemist, which is slow and not scalable to large datasets, and de novo design, which often results in molecules that lack synthetic accessibility and are therefore difficult and costly to pursue. Thus, the aim of this project is to use knowledge-based approaches to propose fragment merges that are synthetically feasible, allowing rapid and cheap progression from fragment hits to lead-like compounds. Various data-driven approaches will be explored, including database exploitation and AI-based techniques, which may be used synergistically to propose new compounds. Use of the former has already been demonstrated during the rotation project, which involved use of the Fragment Network, a graph database containing catalogue compounds, to create a pipeline able to find and filter fragment merges that can be prioritized for further screening. Several avenues to extend this work by enhancing the efficiency of this tool and increasing the diversity in the molecules found have already been identified. Compounds proposed by this project will also have the opportunity for experimental validation. Automated synthesis planning and execution on a robotic platform are currently being explored at XChem, and a key component of this work will be identifying compounds that can be made given the available synthetic repertoire. Integrating this entire pipeline has the potential to be high impact with respect to improving the speed at which we can progress potent ligands in drug development, thereby reducing the number of iterations required for the design-make-test cycle. The proposed work (which will exist in conjunction with another DPhil project, focused on the robotics aspect of the pipeline) will involve industrial collaboration with Vernalis and LifeArc; exact industrial supervisors are to be confirmed. This project falls within the EPSRC's 'Computational and theoretical chemistry' research area and ties strongly to the council's outlined strategies within this field. This research is highly interdisciplinary and will involve collaboration with beamline scientists, medicinal chemists and automation experts. As described above, the software developed during this project will have direct, actionable consequences for the drug discovery community, producing molecules that can be purchased or easily synthesized for further screening. Making source code and data from this project available will allow others to apply this technique to new targets and enable comparison with their own developed algorithms.