Accurate prediction of chemical reactions in solution

点击翻译按钮获取中文摘要

The discovery of new molecules to develop novel materials, agrochemicals, drugs, and therapies is essential to tackle contemporary challenges. However, the rising environmental and economic costs of generating useful novel compounds have placed increased pressure on the chemical sectors. Hence, approaches that can speed up and make the discovery and development processes more efficient are urgently needed. Computational chemistry has become a well-established tool in the development of chemical processes. However, its full potential to transform molecular discovery has been hindered by the limitations of the approaches available. Current methodologies suffer from a lack of generality, accuracy and high cost, making them hard to implement in automated generic workflows. To tackle some of these issues, the Duarte group has developed a computational tool, autodE, which automates the characterisation of reaction pathways using SMILES string representations of reactants and products as inputs. Key features include i) applicability to both organic and organometallic reactions, ii) consideration of conformational sampling of both minima and transition states, and iii) compatibility with several public electronic structure theory packages. However, at present, it is limited by the cost associated with electronic structure methods and solvent description, which hinders its wide use to routinely explore complex catalytic processes and reactions in solution. This is partly due to the description of solvent effects, which is critical in the prediction of chemical reactivity, as solvent effects can bias the preference for a given conformation or reaction pathway. Solvent effects are commonly modelled implicitly with continuum approaches that capture bulk polarisation effects; however, they fail to describe specific solute-solvent interactions, which are crucial to describe charged species. In these cases, explicit solvation may be required to achieve the desired predictive accuracy. The primary objective of this project is to introduce streamlined computational strategies to investigate chemical reactivity in solution, and to use this information to guide reaction optimisation and catalyst design. In collaboration with AstraZeneca, this project seeks to capitalise upon the use of state-of-the-art computational tools and predictive machine learning (ML) models to characterise and predict challenging reactions in solution. More specifically, explicit solvation will be introduced into autodE combining ML potentials for efficient sampling and quadratic string methods to characterise transition states. The implementations developed will be utilised in route design and development and applied for the optimisation of catalysts and substrates in organocatalysed and metal-catalysed reactions. The computational tools arising from this project will be general and widely applicable to different reactions classes and systems, which will accelerate the identification of catalysts and optimal reaction conditions across a wide range of chemical processes. Furthermore, the outcomes of this project will contribute to academic and industrial research in the areas of predictive synthesis, catalysis, process chemistry, computational and theoretical chemistry, and software development. This project falls within the EPSRC Physical Sciences research area.