This project aims to create a new artificially intelligent continuous flow platform for the development of multistep chemical and biocatalysed reactions. Pharmaceuticals are complex molecules which require multiple transformations to synthesise from readily available starting materials. Traditionally they are produced via batch manufacturing, where after each step intermediates are stored in containers or shipped to other facilities around the world to complete the manufacturing process. This adds a significant amount of processing time, contributes to a large carbon footprint, and is at significant risk of supply chain disruptions. In contrast, continuous manufacturing addresses each of these challenges by enabling end-to-end production within the same facility. Catalysts are substances which are added to reactions which influence the rate and/or outcome of the reaction without been consumed. A well-designed catalyst will minimise the generation of waste by being highly selective, recyclable, and only required in very small quantities, often replacing the use of larger amounts of toxic reagents. Hence, it is economically and environmentally desirable to include multiple catalysed steps in a manufacturing process. Alone, the benefits of catalysis and continuous flow are becoming increasingly relevant due to the drive for decarbonisation, but in combination, they have the potential to truly transform the next generation of sustainable manufacturing. However, combining different types of catalysis into continuous flow processes remains highly challenging, due to poor compatibility between catalysts and the large number of variables that need to be optimised. In this project we will develop a fully autonomous and artificially intelligent multistep continuous flow platform, which is capable of simultaneously optimising interconnected catalytic reactions. New multipoint analysis and automated reconfiguration capabilities will enable the creation of individual feedback loops for each reaction, which will be driven by machine learning algorithms suitable for multiobjective and mixed variable systems. We will then demonstrate this approach for the optimisation of industrially relevant chemoenzymatic cascades in sustainable and mutually compatible reaction media (e.g., deep eutectic solvents), thus combining the versatile reactivity of chemocatalysis with the high selectivity of biocatalysis.

Sulfonyl units, that is the -SO2- arrangement of atoms, are functional groups that feature in a significant number of pharmaceuticals, argochemcials and materials. Variations in which one or more oxygen atoms of these groups are replaced with nitrogen atoms are also emerging as useful molecules for exploring biological processes. This proposal is focused on this latter class of compounds. The types of molecules this encompasses - primarily sulfoximines and sulfonimidamides - are less explored than then non-aza-equivalent, and this is largely due to the lack of convenient methods for their preparation. Conventional syntheses of these types of molecules usually involve three or four synthetic operations, and often feature low-yielding steps. The chemistry involved also limits the substrates that can be converted to aza-sulfonyl-containing molecules. This proposal seeks to develop new reagents and new reactions to this class of molecules; the proposed chemistry will be achieved in a single operation, employ readily available reagents and substrates, and be conducted under mild conditions. The key aspect of the proposal is the design of new nitrogen-containing reagents that will allow ready access to a little used class of reactive intermediates; sulfinylnitrenes. By delivering these reactive intermediates in a simple way, using readily available reagents, a host of new reactivity, and thus transformations, will be available. These reactions will be used to provide general routes to sulfoximines and sulfonimidamides, as well as primary sulfonamides. We will deliver user-friendly reactions. These transformations will significantly simplify the preparation of these molecules, and allow them to be routinely considered when new collections of molecules for biological evaluation are being designed. We will seek to make the reagents we develop commercially available, thus allowing the rapid take-up of the methods we develop.

The pharmaceutical industry plays a pivotal role in delivering life-saving medicines to people worldwide. However, the process of making these medicines is often lengthy, costly, and environmentally unsustainable, taking up to 10 years and costing £2Bn and generating up to 100 Kg of waste for every Kg of product. A crucial and ubiquitous step in developing the process of achieving pure, high quality drug substances is crystallisation, where solid drug particles are formed by nucleation and growth from solution. These fundamental process steps are highly unpredictable, sensitive to many parameters and a detailed mechanistic understanding at the molecular scale remains elusive. Developing useful predictive tools to guide the design of this step would have a significant impact with the potential to reduce the cost, time, resources, and waste involved in the design, scale-up and implementation of sustainable manufacturing processes. Current methods used for model-based design of crystallisation processes are not always accurate, failing to capture significant and commonly encountered phenomena such as polymorphism, agglomeration or fouling. This project will change that by blending cutting-edge hybrid machine learning and physics-based computing techniques with our understanding of chemistry and chemical processes. PharmaCrystNet will revolutionise the way we understand and predict crystallisation in drug manufacturing. It aims to: 1) Develop a detailed understanding of the molecular attributes of drug molecules that dictate crystallisation outcomes 2) Develop a new hybrid/ML/mechanistic/physics-informed computer model that can predict crystallisation outcomes under a wide range of industrially relevant process conditions at different scales with high accuracy 3) Test, refine, and validate the model using real-world experiments. This new model will enable: 1) Faster drug production from a 30% reduction in development time, meaning new medicines reach patients more quickly 2) Huge cost savings in the drug manufacturing process, leading to lower drug prices 3) A significant reduction in the environmental footprint of drug production from a 70-80% reduction in material used during development, making the industry more sustainable. By perfecting the crystallisation process, we will propel the pharmaceutical industry into a new era of efficiency and sustainability in generating engineered materials that will deliver further benefits for streamlined efficient downstream drug formulation operations. This project holds promise not just for medicine manufacturers and other specialty chemical manufacturers, but for patients, the environment, and the global community at large.

There is tremendous future scope for biomolecular simulation to provide unprecedented insights into biomolecular systems. The level of detail afforded by these methods, along with their ability to rationalise experimental data and their predictive power are already enabling them to make significant contributions in a wide variety of areas that are crucial for healthcare, quality of life and the environment. The UK biomolecular simulation community has a strong international reputation, with world-leading efforts in in drug design and development, biocatalysis, bionano-technology, chemical biology and medicine. HECBioSim has already delivered outstanding research with impact in bionanotechology, drug design and AMR. But we have only just scratched the surface and there is currently huge room for expansion. Having access to the largest, most modern computing facilities is essential for this. Renewal of the Consortium will enable us to continue allocating time ARCHER for cutting-edge biomolecular simulations. We will place a special emphasis on reaching out to experimentalists and scientists working in industry in order to foster interactions between computational and experimental scientists, and academia and industry to encourage integrated multidisciplinary studies of key problems. Biomolecular simulation and modelling is an integral part of drug design and development. The pharmaceutical industry needs well-trained scientists in this area, as well as the development of new methods (e.g. for prediction of drug binding affinities, ligand selectivity and metabolism). Members of the consortium have a strong track record of collaboration with industry to deliver trained scientists and new methodologies. For example, PhD students trained by consortium members have recently taken up positions in UCB, Unilever, Oxford Nanoimaging and even Sky Broadcasting as software developer. Many of these academic-industry collaborations have been strengthened by work done through HECBioSim allocations. The Consortium will continue to welcome new members from across the whole community. We will continue to develop computational tools and training for both experts and non-experts using biomolecular simulation on HEC resources. We propose to develop new tools that will enable inter-conversion between biomolecular systems at different levels of resolution thereby allowing users to tackle more ambitious 'grand challenges' than are currently feasible. In summary HECBioSim will foster collaborations between computational and experimental scientists between scientists working in industry and academia in all disciplines within biomolecular simulation to maintain the UK as a world-leader in this field.

Development of synthesis and optimisation of reactions remain rate-limiting factors in pharmaceutical process development, often relying on resource-intensive trial-and-error approaches that are costly, time-consuming, and wasteful. This highlights the need to develop new digital methods that are capable of rapidly responding to emerging health challenges. To achieve this, we will create a network of digitally coupled reactors across multiple sites capable of high-throughput screening and self-optimising manufacturing processes. This proposal uniquely combines different flow reactor technologies, analytical techniques, and automated workflows to provide enhanced mapping of chemical space and generation of robust high-quality datasets. Robotics will be used to design flexible experimental systems capable of exploring continuous (e.g., time, temperature) and categorical (e.g., catalyst, ligand) variables, as well as different reactor types. Notably, parallelised droplet flow reactors will be developed and combined with intelligent optimisation algorithms to reduce the amount of material required during pharmaceutical development campaigns. A multisite reactor network will be established and driven by next generation machine learning algorithms, which will use knowledge from prior experimental campaigns to increase library synthesis success rates and accelerate the development and optimisation of chemically related processes. Orders of magnitude more experiments are performed during discovery than during process development; the high-quality automated data collected at this early stage will be essential for accelerated, lower cost and sustainable manufacturing. In collaboration with our partners in the pharmaceutical industry, we will leverage this novel workflow to streamline the pathway to future medicines. The capabilities and results generated from our delocalised artificially intelligent network will be transferable across different chemical manufacturing sectors. The objectives of this research are: Development of autonomous high-throughput microfluidic flow reactors for the synthesis of pharmaceutically relevant compound libraries. Library synthesis success rates will be increased by integration of state-of-the-art mixed variable optimisation algorithms. Real-time online analytics will be used to quantify each reaction, thus providing robust and standardised datasets for use in predictive machine learning models, enabling their application towards currently underexplored chemistries. Creation of digitally coupled reactors across multiple sites for the exploration of wide process spaces. To achieve this, complementary analytical techniques and different reactor technologies will be leveraged to generate datasets across different scales. Parallelised optimisations will consider the trade-offs between multiple objectives, enabling the sustainability of manufacturing to be considered from the outset of pharmaceutical development. Combination of different types of data across multiple experimental labs to generate hypotheses for new library synthesis and process optimisation campaigns. Next generation machine learning algorithms will be designed to use prior knowledge of contextually similar chemical systems, with the aim of accelerating the transition from discovery to manufacturing. Demonstration of a pilot-scale manufacturing process. Our network of digitally coupled reactors will be used to perform parallelised library synthesis and self-optimisation of a selected process. Scale-up will be evaluated using the facilities available within the iPRD at Leeds.