Conclusions of the 7th Crystal Structure Prediction Blind Test
Crystal Structure Prediction boundaries pushed beyond the pharmaceutical sector to areas such as electronics and photonics.
- Seven structures of 2D systems that had been experimentally analysed but not published (‘blind’) released to participants along with some experimental conditions.
- Boundaries pushed beyond the pharmaceutical sector to areas such as electronics and photonics with the inclusion of an optoelectronic system containing silicon and a metal-organic system containing copper.
- Also included one of the most challenging systems in CSP Blind Test history - a large, highly polymorphic, phamaceutical drug candidate.
- Test ran from October 2020 to September 2022.
- More experimental data released over the course of the test to simulate real world conditions.
- Submissions received from industrial and academic groups worldwide.
- All compounds successfully had their experimentally observed crystal structures predicted from landscapes of 1K+ potential structures.
- PXRD challenge showcased the application of CSP to industrially relevant molecules.
The 7th Crystal Structure Prediction (CSP) Blind Test has recently
closed.
CSP is the ability to predict, from its 2D molecular structure, the 3D crystal
structure(s) that a given molecule will form. Techniques used include informatics and
computational science combined with intensive computational resources.
If 3D crystal structures could be accurately and consistently predicted from 2D
drawings, stability risks could be predicted before costly experimental trials. This
would be highly advantageous when developing pharmaceuticals and other solid form
materials.
The search for the experimentally observed crystal structure begins with
building a molecular model from a 2D representation of a molecule. Advanced search
techniques are then used to generate plausible crystal structures that can then be
visualized and ranked on energy and density.
However, which crystal structure(s) occur experimentally is not that simple.
The occurrence of many factors, including stable and meta stable polymorphs, packing
complications, and the fact that the lowest energy structure is not always
experimentally observed, presents CSP with many challenges. The CCDC’s Blind Tests,
which began in 1999, bring together scientists in this field to advance methods and
overcome these and other challenges.
In the 7th CSP Blind Test that ran from October 2020 to September 2022, seven
structures of 2D systems that had been experimentally analysed but not published
(‘blind’) were released to participants along with some experimental
conditions.
The seven 2D structures included Cu and Si-containing systems which pushed the
boundaries of CSP beyond the pharmaceutical sector to areas such as electronics and
photonics.
Other structures included one of the most challenging systems in CSP Blind Test
history – a large, highly polymorphic, pharmaceutical drug candidate – along
with agrochemicals and a food flavouring.
More experimental data was released over the course of the test to simulate
real world conditions.
Participants included groups from both industry and academia, and all compounds
successfully had their experimentally observed crystal structures predicted by at least
one group from landscapes of 1K+ potential structures.
In one challenge, a PXRD pattern for an observed crystal structure was provided
alongside the 2D chemical structure. This emulated a common situation where a single
crystal structure is unable to be obtained, but a poor-quality powder pattern is. This
showcased an application of CSP to industrially relevant molecules.
CSP is an ever-evolving discipline and the 7th Blind Test identified future
challenges including disorder prediction that had previously been disregarded owing to
its complexity but is now more industrially relevant to solve as bigger molecules with
more flexible groups are being seen in drug development; more challenges that reflect
industrial reality; and continued broadening beyond pharmaceuticals to solid-state
devices.
“Thanks to collaborative initiatives like the CCDC blind test, the
complex field of CSP is advancing rapidly, taking advantage of machine learning
techniques and the availability of ever-more powerful computing resource,”
says Dr Jürgen Harter, CEO, CCDC. “I look forward to future Blind
Tests addressing challenges such as structure ranking, overprediction and
disorder.”
A scientific paper with full results is being prepared and will be submitted
for publication in 2023. Preliminary detailed results, including how many groups
correctly predicted each target, the lowest rankings, CPU time used and details of
methodologies available by emailing us.