Improving Crystal Structure Prediction via Tailored Force Fields
Lead Research Organisation:
Imperial College London
Department Name: Chemical Engineering
Abstract
There has been tremendous progress in recent years in the development of practical approaches to crystal structure prediction (CSP) for polymorphic molecules and co-crystal systems. It is now possible to predict the crystal energy landscape of molecules of practical agrochemical or pharmaceutical interest, with multiple flexible torsions and bond angles. CSP studies can now support decision making and risk management, either by helping to design experiments aimed at crystallising a specific structure or by offering re-assurance that all likely polymorphs have already been identified. Given a sufficiently accurate model, such studies can also form the basis for the prediction of numerous crystal and solution properties.
Current state-of-the-art techniques, such as those incorporated in the CrystalPredictor and CrystalOptimizer codes developed by the groups of Claire Adjiman and Costas Pantelides, generally assume that the only available information is the molecular structure. Although this assumption significantly widens the applicability of the approach, it is often unnecessarily restrictive. In particular, one or more crystal structures for the compound of interest may already be already known or may be identified during the course of active ingredient development. To fully realise the potential of CSP techniques within an industrial context, it is important to make use of all such available information.
The aim of this project is to develop a CSP methodology that can take advantage of prior knowledge in order to achieve greater accuracy and reliability. The use of such information to accelerate the search for polymorphs will also be investigated.
The proposed project will build on recent work within the Molecular Systems Engineering group at Imperial College that has highlighted the importance of the empirical repulsion/dispersion terms used in many of the models at the heart of CSP. This has led to techniques and codes for the derivation of improved parameter values using available experimental structures. Overall, this has been found to lead to much more accurate lattice energy calculations. However, this work has been focused on rigid molecules, which are simpler than the more flexible molecules typical of the agrochemicals industry.
The primary objective of the proposed project will be develop a formal framework for determining force fields that are tailored to specific compounds of interest. Our proposed approach will combine ab initio CSP techniques with prior solid state information that is available for these or closely-related compounds. The effectiveness of the new methodology will be assessed via testing on several compounds of industrial relevance.
Current state-of-the-art techniques, such as those incorporated in the CrystalPredictor and CrystalOptimizer codes developed by the groups of Claire Adjiman and Costas Pantelides, generally assume that the only available information is the molecular structure. Although this assumption significantly widens the applicability of the approach, it is often unnecessarily restrictive. In particular, one or more crystal structures for the compound of interest may already be already known or may be identified during the course of active ingredient development. To fully realise the potential of CSP techniques within an industrial context, it is important to make use of all such available information.
The aim of this project is to develop a CSP methodology that can take advantage of prior knowledge in order to achieve greater accuracy and reliability. The use of such information to accelerate the search for polymorphs will also be investigated.
The proposed project will build on recent work within the Molecular Systems Engineering group at Imperial College that has highlighted the importance of the empirical repulsion/dispersion terms used in many of the models at the heart of CSP. This has led to techniques and codes for the derivation of improved parameter values using available experimental structures. Overall, this has been found to lead to much more accurate lattice energy calculations. However, this work has been focused on rigid molecules, which are simpler than the more flexible molecules typical of the agrochemicals industry.
The primary objective of the proposed project will be develop a formal framework for determining force fields that are tailored to specific compounds of interest. Our proposed approach will combine ab initio CSP techniques with prior solid state information that is available for these or closely-related compounds. The effectiveness of the new methodology will be assessed via testing on several compounds of industrial relevance.
Studentship Projects
| Project Reference | Relationship | Related To | Start | End | Student Name |
|---|---|---|---|---|---|
| EP/N509486/1 | 01/10/2016 | 31/03/2022 | |||
| 1966646 | Studentship | EP/N509486/1 | 01/10/2017 | 31/03/2021 | David Bowskill |
| Description | So far, this award has help develop an understanding of key routes to improve the reliability of cost-effective tailored force fields, and highlighting the importance of reparameterisation of repulsion/dispersion type interactions and their impact on determining crystal polymorph stabilities. Persuing this, advanced codes have been developed to reparameterise such potentials based on available reference data. In previous work, reference data has been obtained from experimental sources, however a review done as a result of this award has demonstrated that such data is unreliable and often sporadic. This has led to the careful construction of large sets of reference data obtained from rigorous and costly theoretical quantum mechanical calculations (Periodic DFT-D) as an alternative to experimental data. Such reference data is expected to be used repeatedly throughout the the remainer for the project, and beyond it, as well as providing useful benchmark calculations for the wider community. Initial results have shown that the use of theorical reference data along with key algorithmic developments to reparameterise the repulsion/dispersion potentials can lead to significant improvements in the capability of Crystal Structure Prediction investigations at a modest increase in computational cost when compared to competing methodologies. This has been demonstrated in a recently accepted publication: "Efficient parameterisation of a surrogate model of molecular interactions in crystals" in the proceedings of the 30th European symposium on computer aided process engineering (2020). Additional work has since been carried out to construct a highly efficient code for the optimisation of crystal structures using both point charge based and the more complex multipole based electrostatic models. The developed code is known as Crystal Structure Optimizer - Rigid Molecules (CSO-RM) and, to my knowledge, is the first code of its type to combine several elements such as structure optimisation, analytical second derivative calculations, and multipole energy functions. Through methodological developments the speed of this code is equivalent or better than compariable methodologies such as the code DMACRYS developed at UCL which performs many of the same functions but without analytical second derivative calculations. CSO-RM is also a reusable software component and as such has been incorporated into the parameter estimation methodology and with algorithmic improvements this reduces the computational cost of parameter estimation approaches by approximately 3 orders of magnitude while achieving unprecidented accuracy compared to previous approaches. This has been tested predominantly in the development of transferable potentials for crystal structure prediction by fitting to the theoretical quantum mechanical calculations (Periodic DFT-D) that were also performed as part of this work. Testing of these parameters against popular literature potentials shows remarkable improvement in the prediction of both crystalline energies and geometrys, as well as more reliable results from CSP investigations. Further work is currently being carried out on testing these potential further and publications are in preparation. Resources supplied by this project have also facilitated improved collaboration with experimental groups as demonstrated in the publication "Efficient screening for ternary molecular ionic cocrystals using a complementary mechanosynthesis and computational structure prediction approach." published in Chemistry-A European Journal (2019), "Can solvated intermediates inform us about nucleation pathways? The case of ß-pABA" published in CrystEngComm (2020), and "Three new polymorphs of 1,8-diacetylpyrene: a material with packing-dependent luminescence properties and a testbed for crystal structure prediction" published in Journal of Materials Chemistry (2021) |
| Exploitation Route | Outcomes for this work will lead to the publishing of tranferable parameter sets with improved reliability over existing parameter sets in literature. It is hoped that these will see common use in both academic and industrial applications. Codes developed during this award which include recent algorithmic developments will also be supplied to the community to advance the development of intermolecular potentials in crystal structure prediction. Common industrial applications include pharmaceutical drug discovery, and the development of agrochemicals and organic semiconductors. |
| Sectors | Agriculture, Food and Drink,Chemicals,Electronics,Pharmaceuticals and Medical Biotechnology |
| URL | https://onlinelibrary.wiley.com/doi/abs/10.1002/chem.201904672;https://pubs.rsc.org/fa/content/articlehtml/2020/ce/d0ce00970a;https://pubs.rsc.org/en/content/articlehtml/2021/tc/d0tc05522k |
| Description | Kalifa University |
| Organisation | Khalifa University |
| Country | United Arab Emirates |
| Sector | Academic/University |
| PI Contribution | This collabortion supported a large scale experimental investigation with computational calculations. A colleague in my research team performed multiple CSP investigations on systems of interest to the experimental group while I performed theortical calculations to assess thermodynamic driving forces in the formation of ionic cocrystals, and thus, help to rationalise experimental observations. |
| Collaborator Contribution | Our partners performed experimental studies with our computational results help to support. |
| Impact | 10.1002/chem.201904672 This collaboration primarily involved groups from Chemistry and Chemcial Engineering backgrounds. |
| Start Year | 2019 |