processing, structure solution and refinement and integration with follow-on (data
science) methods. The automation of data acquisition and structure solution
discussed herein is a clear step in this direction that has largely been taken already.
However, these approaches do not generally work for more complicated and difficult
cases.
Better software and algorithms for data processing will be crucial for the future of
the discipline. We have highlighted above that there is some work to be done as
modern detectors evolve. However, current focus is on static structures and hence
getting the best data integration for Bragg peaks – and in a significant number of
cases, this is a simplistic view of the actual crystalline state. The ability to understand
the total scatting pattern reveals the full behaviour in the crystal and provides great
insights. This is not currently an easy or routine approach to take, and addressing this
issue would open up a myriad of new structural chemistry. Routine analysis of the
total scattering pattern would enable true modelling of ‘disorder’ and better analysis
of local structure in complex materials and would provide the basis for dynamic
crystallography methods to really thrive.
These advances in data processing call for advances in structure refinement. It
will be necessary to develop new approaches to disorder modelling, and these could
also be augmented by a closer operation with the databases. If refinement software
worked more in tandem with the databases, it would be possible to learn from, and
use, models (or parts of models) that already exist. A full and automated integration
between the two would ensure that as the databases grow, they could increase in
quality. Machine learning methods are now beginning to gain a lot of traction across
many data-driven research areas – they have the potential to make structure refinements better, to make database records better and to power entirely new research in
structural chemistry and in linking to other areas of science. There are also clear
advantages in the convergence of experimental crystallography with Crystal Structure Prediction – the seamless interplay of these two approaches would mean that
many more insights into an experimental structure would readily be possible and
could feed into experiment design as well as interpretation.
4.2.3 Data
A Data Infrastructure
For a fully integrated and end-to-end software infrastructure to be realised, there is
the necessity for the parallel development of a better, complementary data infrastructure. The crystallographic community has pioneered in many respects through
the development of the Crystallographic Information Framework, and this provides
the basis for a twenty-first-century data infrastructure. Modern automation
approaches are a good example of leveraging data and metadata standards alongside
software development. However, there are several areas where the data infrastructure
clearly needs to be extended.
128
S. J. Coles et al.
science) methods. The automation of data acquisition and structure solution
discussed herein is a clear step in this direction that has largely been taken already.
However, these approaches do not generally work for more complicated and difficult
cases.
Better software and algorithms for data processing will be crucial for the future of
the discipline. We have highlighted above that there is some work to be done as
modern detectors evolve. However, current focus is on static structures and hence
getting the best data integration for Bragg peaks – and in a significant number of
cases, this is a simplistic view of the actual crystalline state. The ability to understand
the total scatting pattern reveals the full behaviour in the crystal and provides great
insights. This is not currently an easy or routine approach to take, and addressing this
issue would open up a myriad of new structural chemistry. Routine analysis of the
total scattering pattern would enable true modelling of ‘disorder’ and better analysis
of local structure in complex materials and would provide the basis for dynamic
crystallography methods to really thrive.
These advances in data processing call for advances in structure refinement. It
will be necessary to develop new approaches to disorder modelling, and these could
also be augmented by a closer operation with the databases. If refinement software
worked more in tandem with the databases, it would be possible to learn from, and
use, models (or parts of models) that already exist. A full and automated integration
between the two would ensure that as the databases grow, they could increase in
quality. Machine learning methods are now beginning to gain a lot of traction across
many data-driven research areas – they have the potential to make structure refinements better, to make database records better and to power entirely new research in
structural chemistry and in linking to other areas of science. There are also clear
advantages in the convergence of experimental crystallography with Crystal Structure Prediction – the seamless interplay of these two approaches would mean that
many more insights into an experimental structure would readily be possible and
could feed into experiment design as well as interpretation.
4.2.3 Data
A Data Infrastructure
For a fully integrated and end-to-end software infrastructure to be realised, there is
the necessity for the parallel development of a better, complementary data infrastructure. The crystallographic community has pioneered in many respects through
the development of the Crystallographic Information Framework, and this provides
the basis for a twenty-first-century data infrastructure. Modern automation
approaches are a good example of leveraging data and metadata standards alongside
software development. However, there are several areas where the data infrastructure
clearly needs to be extended.
128
S. J. Coles et al.