The Cambridge Crystallographic Data Centre (CCDC) announces two significant scientific papers detailing the findings of the 7th Crystal Structure Prediction (CSP) Blind Test. To coincide with the publication of the two papers, the CCDC has released a CSP Blind Test database that contains 171,679 entries from 207 landscapes.
Expanding CSP Beyond Pharmaceuticals
CSP predicts the most likely crystal structures to form from a given molecule, based on its 2D chemical structure. Most methods use informatics and computational science techniques. Predicting more stable structures leads to many benefits including improved manufacturing processes, patent protection and breaking, and the potential discovery of new, improved materials.
Since 1999 the Crystal Structure Prediction (CSP) Blind Tests have brought together scientists from industry and academia to test their methods against a real example in a controlled environment and advance methods.
The 7th CSP Blind Test, which ran from October 2020 to September 2022, involved the analysis of seven 2D chemical systems. These included systems containing copper and silicon, pushing the boundaries of CSP beyond pharmaceuticals to areas such as electronics and photonics. The test also featured one of the most challenging systems in CSP history—a large, highly polymorphic pharmaceutical drug candidate—along with agrochemicals and a food flavouring.
Participants attempted to predict the experimentally observed crystal structures for all compounds from landscapes of over 1,000 potential structures. A PXRD-based challenge demonstrated the use of CSP to solve a crystal structure from a powder pattern where single crystal structures are unattainable - a situation commonly encountered for pharmaceuticals.
Two Papers in Acta Crystallographica
These papers offer comprehensive insights into the latest advancements in crystal structure prediction, a field crucial for the development of pharmaceuticals, electronics, and photonics.
Scientific Paper 1: The Seventh Blind Test of Crystal Structure Prediction: Structure Generation Methods
The first paper (Hunnisett et al., J. Acta. Cryst. B80, Dec 2024) covers the methodologies and results of generating crystal structures for seven target systems of varying complexity.
Key findings from this study include:
- Successful reproduction of experimentally observed crystal structures for a small but flexible agrochemical compound by many CSP methods
- Limited success for systems of higher complexity, highlighting ongoing challenges in the field
- Demonstration of CSP in determining crystal structures from low-quality powder X-ray diffraction (PXRD) patterns
- Exploration of CSP in predicting likely co-crystal stoichiometry, showcasing multiple potential approaches
- Emergence of crystallographic disorder as a significant theme, with two groups successfully predicting the existence of disorder for the first time
- Large-scale comparisons of predicted crystal structures, revealing that some methods yield largely similar sets of structures
Scientific Paper 2: The Seventh Blind Test of Crystal Structure Prediction: Structure Ranking Methods
The second paper (Hunnisett et al., J. Acta. Cryst., Sect. B, in press) focuses on the methods used to rank crystal structures based on their stability. This part of the test involved standardized sets of structures generated from a variety of methods. Key participants from 22 groups applied various approaches, including:
- Periodic DFT-D methods
- Machine-learned potentials
- Force fields derived from empirical data or quantum chemical calculations
- Combinations of the above methods
- One non-energy-based scoring function
Significant outcomes of this study include:
- General agreement of periodic DFT-D methods with experimental data within expected error margins
- Promising results from a machine-learned model using system-specific AIMnet potentials, suggesting an efficient alternative to DFT-based methods
- Consensus across periodic DFT methods for one target, indicating that a more stable polymorph may exist but has not yet been observed
- Variable improvements to predictions - i.e. free energies didn’t always provide an improvement
- Recommendations for future research to enhance prediction efficiency due to the vast resources utilized in many cases
Future Challenges and Directions
The 7th Blind Test highlighted several future challenges, including the prediction of crystallographic disorder and the need for more realistic industrial scenarios. The field is rapidly advancing with the integration of machine learning techniques and powerful computational resources.
“This test has showcased some extraordinary gains in the field of crystal structure prediction. The increasing intricacies, intelligence and sophistication of some of these methods are just remarkable. I am incredibly grateful to have led such a legendary initiative and thank all my colleagues, both at the CCDC and across participant groups, for their invaluable efforts and contributions to this project, “ said Dr Lily Hunniset, CCDC Blind Test 7 lead and principal author of the 2 research papers.
“These two new peer-reviewed scientific papers are the culmination of another scientifically challenging crystal structure blind test that once again has pushed the boundaries, overcome previous challenges, and opened up new opportunities in this vital discipline in the development of new materials, including pharmaceuticals for the benefit of us all. I take this opportunity to thank all the CSP Blind Test 7 participants across the globe for their dedication and their enquiring and innovative solutions to the CSP challenges set by the test,” Jürgen Harter, CEO CCDC.
