Synthesis and Structure Elucidation of Natural Products

在这个项目中，我们的目标是自动制备聚酮化合物，这样机器人就可以在很少或没有人为干预的情况下制作整个聚酮化合物库。因为这个项目将允许我们在短时间内访问数十或数百个聚酮，我们也将开发基于量子力学和机器学习的非常快速的计算工具来设计聚酮，然后在它们制造出来后对它们进行分析。聚酮类化合物是一类天然存在的化学物质，占目前最畅销药物的20%左右，包括抗生素、抗真菌药和抗肿瘤药。因此，它们对人类健康至关重要，它们的制备（称为“化学合成”的过程）是20世纪的主要成功之一。然而，这些分子的结构可能极其复杂——它们的化学合成极具挑战性，事实上，每种聚酮的合成都是一个单独的、定制的科学研究，需要数月、数年或数十年的工作才能完成。即使它们被制造出来，它们的复杂性也使得研究它们的结构和行为成为另一项具有挑战性的任务——例如，如果我们希望了解它们的三维结构和运动，那么我们通常不得不依赖苛刻的量子化学计算，这需要几天/几周/几个月的高端计算时间来进行。所有这些都与其他重要的自然产生的分子形成对比，如肽（构成体内蛋白质）或DNA。对于这些分子，化学合成现在完全是常规的，可以完全自动化——科学家只需拨通他们想要的化合物，机器人就可以用简单、现成的积木来构建它。我们想要让那些极具价值的聚酮类化合物变得更容易获得。我们已经开发了制造聚酮化合物成分的方法，现在寻求在我们新收购的Chemspeed自动化平台（机器人化学合成仪器）上自动化组装。通过在机器人上组合不同的构建模块，可以快速获取多种复杂的聚酮，这反过来将使生物学研究能够了解结构如何影响生物活性。制造这么多新的、复杂的化合物的缺点是，在设计或分析我们想要制造的分子结构时，尤其是它们的三维结构，会遇到瓶颈。我们目前最先进的量子化学方法非常缓慢（但难以置信的准确），根本无法跟上。因此，我们建议在我们最近开发的超高速机器学习系统的基础上，模拟量子化学计算，但在毫秒内而不是几天或几个月。我们制造的机器人合成和由此产生的化合物将使我们能够开发和测试更精确的量子化学方法，然后使用这些方法大规模改进我们的机器学习系统，使其足以快速筛选我们可能合成的数百甚至数千种潜在结构。有了制造聚酮的自动合成机器人和让我们设计和研究这些化合物的机器学习系统，我们最终的目标是使聚酮合成像肽和DNA合成一样简单——彻底改变这些分子的开发方式。

In this project we will aim to automate the preparation of chemical compounds called polyketides, so that a robot can be programmed to make an entire library of polyketide compounds with little or no human intervention. Because this project will allow us to access dozens or hundreds of polyketides in a short space of time, we will also be developing very fast computational tools based on quantum mechanics and machine learning to design polyketides and then analyse them once they are made.Polyketides are a class of naturally occurring chemicals that comprise around 20% of the current top-selling drugs, including antibiotics, antifungals and anti-tumour agents. As such they are crucially important to human health and preparation of them (a process called "chemical synthesis") has been one of the major successes of the 20th century. However, the structures of these molecules can be extremely complex - and their chemical synthesis is incredibly challenging, in fact the synthesis of each polyketide is a separate, bespoke scientific investigation taking months, years or decades of work to complete. Even when they are made, their complexity makes the study of their structures and behaviours another challenging task - for example if we wish to understand their 3-dimensional structure and motion, then we often have to rely on demanding quantum chemical calculations that require days/weeks/months of high-end computing time to undertake. All of this contrasts with other important naturally occurring molecules, such as peptides (which make up the proteins in the body) or DNA. For these molecules, the chemical synthesis is now totally routine and can be fully automated - the scientist simply dials-up the compound they want and the robot can construct it from simple, readily available building blocks. We want to make the incredibly valuable polyketide class of compounds just as accessible.We have developed methodology for making the components of polyketides and now seek to automate their assembly on our newly acquired Chemspeed Automated Platform - a robotic chemical synthesis instrument. Through combining different building blocks on the robot, a diverse set of complex polyketides can be rapidly accessed, which in turn will enable biological studies to see if how the structure affects the biological activity. The downside of making so many new, complex chemical compounds, is then the bottle-neck created in designing or analysing the structures of the molecules we would like to make - in particular their three-dimensional structures. Our current state-of-the-art quantum chemical approaches to this are very slow (but incredibly accurate) and will simply not be able to keep up. So we propose to build on our recent development of an ultra-fast machine learning system that can mimic quantum chemical calculations, but in milliseconds rather than days or months. The robotic syntheses and resulting compounds that we make will allow us to develop and test more accurate quantum chemical methods and then use these to massively improve our machine learning system so that it is good enough to rapidly screen the hundreds or even thousands of potential structures that we might synthesise. With both the automated synthesis robot to make polyketides and the machine learning system that lets us design and study these compounds we ultimately aim to render polyketide synthesis as easy as peptide and DNA synthesis - revolutionising the way that these molecules are developed.