Periodic Reporting for period 4 - CoMMaD (Computational Molecular Materials Discovery)
Période du rapport: 2022-10-01 au 2023-09-30
In the simplest of definitions, chemistry concerns the synthesis and the properties of molecules. Supramolecular chemistry is known as “chemistry beyond the molecule”, where groups of molecules assemble without forming chemical bonds. Chemists and materials scientists aim to make these “molecular materials” or “supramolecular systems” have useful properties for a wide range of potential useful applications. Supramolecular systems have exciting applications as hosts for guests, sensors, molecular switches, performing molecular sieving and as catalysts that speed up other reactions. While the potential applications of these systems are broad, and some may be those we do not currently expect, improved materials for molecular separations, or for the generation of renewable energy can help in our drive towards meeting the EU 2030 Energy Strategy of a 40% reduction in greenhouse gas emissions and at least 27% renewable energy. The overarching goal of this project is to accelerate the discovery process of these types of molecular materials through the use of computation.
We would like to design molecular materials systems for new applications by deducing the properties of a system from a simple chemical sketch or idea – much as an architect’s sketch of a building, for example, can reliably predict its function. However, when we simply draw a molecule, we do not know what properties it will have, nor how it will assemble. Worse, in many cases we cannot be confident that the particular molecule can in fact be synthesised at all since the assembly rules in chemistry are, still, much less well developed than those in architecture. Instead, synthetic chemists use their chemical intuition to guide them as to the best experiments to try. Then, if successful in getting a product, they must characterise the material and its properties. Even in state-of-the-art labs, this is a slow process – a new molecule can take a year to prepare, let alone to characterise. Sometimes even small changes in the reaction can have a large effect on the outcomes, hence ‘intuitive’ design breaks down, particularly as systems become more complex.
My aim in this project is to provide a computational ‘blueprint’ for molecular materials in order to allow synthetic research teams to discover new, targeted functions in a much more rapid timeframe. We can predict firstly which molecules form and then how they will assemble. This is exciting because it will allow us to direct chemists towards the best synthetic systems and my overarching goal is to show that computational modelling can be responsible for the discovery of new materials with useful new applications, rather than simply rationalising results from synthetic teams.
We would like to design molecular materials systems for new applications by deducing the properties of a system from a simple chemical sketch or idea – much as an architect’s sketch of a building, for example, can reliably predict its function. However, when we simply draw a molecule, we do not know what properties it will have, nor how it will assemble. Worse, in many cases we cannot be confident that the particular molecule can in fact be synthesised at all since the assembly rules in chemistry are, still, much less well developed than those in architecture. Instead, synthetic chemists use their chemical intuition to guide them as to the best experiments to try. Then, if successful in getting a product, they must characterise the material and its properties. Even in state-of-the-art labs, this is a slow process – a new molecule can take a year to prepare, let alone to characterise. Sometimes even small changes in the reaction can have a large effect on the outcomes, hence ‘intuitive’ design breaks down, particularly as systems become more complex.
My aim in this project is to provide a computational ‘blueprint’ for molecular materials in order to allow synthetic research teams to discover new, targeted functions in a much more rapid timeframe. We can predict firstly which molecules form and then how they will assemble. This is exciting because it will allow us to direct chemists towards the best synthetic systems and my overarching goal is to show that computational modelling can be responsible for the discovery of new materials with useful new applications, rather than simply rationalising results from synthetic teams.
During the final work period of the project, that is from 1st Oct 2022 to 30th Sept 2023, significant progress was made in the development and application of a computational approach to accelerate the discovery of molecular materials. Firstly, we have continued to develop open-source discovery software, in particular the supramolecular toolkit (stk), and shown that it can apply an evolutionary algorithm to explore the possible chemical space of molecular materials across several different systems. We have disseminated our approach across several reviews and book chapters, and are aware of several international groups, some experimental, now routinely using our software prior to material synthesis. We also met milestones relating to machine learning materials' properties and the experimental realisation of some of our predictions.
We have particularly made progress in the development of software for porous materials, with many publications relating to that topic, including one in Science for polymer membranes. We have further developed new approaches for the structure prediction of porous (and potentially other) systems using coarse-graining approaches which has already results in several publications and presentations at international conferences. Finally, we have progressed well in developing our software into a new area of organic electronics, with several publications already and further expected in the near future.
We also completed work on developing an approach to consider whether the hypothetical materials that we predict to have promising properties can actually be made in the laboratory. We developed an approach using machine learning to do this, where our algorithm can analyse the molecular building blocks of a hypothetical material and “predict” whether our synthetic collaborators would say the material is something they believe they could make in the experimental laboratory. We have also built libraries of hypothetical materials that are open-source along with our open-source software and publications, and presentation at many international conferences.
We have particularly made progress in the development of software for porous materials, with many publications relating to that topic, including one in Science for polymer membranes. We have further developed new approaches for the structure prediction of porous (and potentially other) systems using coarse-graining approaches which has already results in several publications and presentations at international conferences. Finally, we have progressed well in developing our software into a new area of organic electronics, with several publications already and further expected in the near future.
We also completed work on developing an approach to consider whether the hypothetical materials that we predict to have promising properties can actually be made in the laboratory. We developed an approach using machine learning to do this, where our algorithm can analyse the molecular building blocks of a hypothetical material and “predict” whether our synthetic collaborators would say the material is something they believe they could make in the experimental laboratory. We have also built libraries of hypothetical materials that are open-source along with our open-source software and publications, and presentation at many international conferences.
We have made developments beyond the state of the art in multiple areas, key highlights include:
• Our open-source software, the supramolecular toolkit (stk), has been extended to include an evolutionary algorithm, mimicking “survival of the fittest” in nature in order to efficiently explore the almost infinite number of possible molecular materials that could hypothetically be possible. The software is able to assemble and assess an increasing number of different molecular materials, including porous organic cages in more than twenty different structures, metal-organic cages, rotaxanes and other interlocked molecules, small molecules of multiple components and linear polymers. Also extended and applied to organic electronic systems, with some publications in this are and further to come.
• We have built databases of tens of thousands of porous molecules, assessed for their properties. Further, we have developed the first machine learning (artificial intelligence) algorithm that predicts their key properties, instead of an experimental chemist having to work for 1-2 years in the laboratory to synthesise the molecule and determine its properties.
• We have demonstrated for porous molecules for the first time that, starting only from knowledge of the 2-dimensional structure of the precursor molecules, it is possible to firstly predict which molecules will form from a range of possibilities, and then to predict how those molecules will assemble in the solid-state.
• We have demonstrated the ability to assemble approx 350,000 linear polymers models and rapidly assess their properties, compared to a handful of experimental reports.
• Our polymer membrane simulations have assisted in the discovery of new polymers with beyond state of the art properties for a range of applications, including molecular sieving and in flow batteries.
• We have developed an approach using a machine learning model that assesses whether material precursors are likely to be synthesisable or not.
• Our open-source software, the supramolecular toolkit (stk), has been extended to include an evolutionary algorithm, mimicking “survival of the fittest” in nature in order to efficiently explore the almost infinite number of possible molecular materials that could hypothetically be possible. The software is able to assemble and assess an increasing number of different molecular materials, including porous organic cages in more than twenty different structures, metal-organic cages, rotaxanes and other interlocked molecules, small molecules of multiple components and linear polymers. Also extended and applied to organic electronic systems, with some publications in this are and further to come.
• We have built databases of tens of thousands of porous molecules, assessed for their properties. Further, we have developed the first machine learning (artificial intelligence) algorithm that predicts their key properties, instead of an experimental chemist having to work for 1-2 years in the laboratory to synthesise the molecule and determine its properties.
• We have demonstrated for porous molecules for the first time that, starting only from knowledge of the 2-dimensional structure of the precursor molecules, it is possible to firstly predict which molecules will form from a range of possibilities, and then to predict how those molecules will assemble in the solid-state.
• We have demonstrated the ability to assemble approx 350,000 linear polymers models and rapidly assess their properties, compared to a handful of experimental reports.
• Our polymer membrane simulations have assisted in the discovery of new polymers with beyond state of the art properties for a range of applications, including molecular sieving and in flow batteries.
• We have developed an approach using a machine learning model that assesses whether material precursors are likely to be synthesisable or not.