data quality. The chemical crystallographer will often ‘get what they are given’ with
limited opportunity to grow better crystals, whereas the macromolecular crystallographer will robotically attempt to grow crystals of the same sample in highthroughput modes under as many different conditions as possible.
These factors naturally lead to a much wider range, in terms of the quality, of
crystal structures produced by modern facilities. However, the ability to validate
results to a greater extent has also progressed. Due to the increased standardisation
and digitisation of data and results, there are now not only a larger number of
quantifiable metrics to assess quality, but also they can be combined and compared
to further probe and measure quality. A couple of decades ago, a structure would be
judged by just a handful of metrics, and often the non-expert would only consider the
R-factor. However, modern validation tools are now accessible to all and results
presented in such a way that it is relatively easy to judge quality based on a whole
range of factors. So, one might now consider that it would be possible to publish a
much wider range of quality of results, if not all that can be deemed as ‘correct’. This
is a crucial change in the mindset of publishing, but a key one to support modern
chemistry requirements.
This gives rise to the concept that ‘data need to be fit for purpose’. The primary
purpose of a service crystallography structure in the past was essentially a proof of
what has been synthesised. Increasingly, over the years, crystallography has moved
on from this configuration/conformation requirement to look in detail at the nature of
bonding and crystal structures in the form of intermolecular interactions and thus
requiring a certain quality of result. So, it should be perfectly acceptable, for
example, to publish a structure that critically reveals the stereochemistry of a
compound but does not reliably provide any detailed insight into bonding. Of course,
the structure must be correct, but it can be of low quality – and these factors can
readily be assessed.
Furthermore, in the modern era, it is now critically important to consider that a
result has a very (in some cases, more?) important role when incorporated in and
contributing to the whole body of knowledge, e.g. a structure becoming a component
of a database. There are illustrations throughout the literature, and this article, of the
benefits of being able to analyse a database to derive rules and provide the basis for
informatics. In an ideal world therefore, every structure determined should in
principle contribute to this body of knowledge.
We can make a big step towards this goal. In the modern world of data science, it
is very common to spend a significant amount of time ‘cleaning’ and ‘processing’ a
large dataset. If we are to move towards a ‘data-fit-for-purpose’ approach, then
validation becomes an imperative process. This can readily be achieved not only
using our modern validation tools but also by making all of our data available. In
recent years common practice has shifted, over a relatively short period of time, from
provision of structure factors being very rare to this being close to mandatory for
most publishing routes. Deposition of structure factors has empowered a whole new
level of validation, where not only the model can be compared to the data from
which it was derived but also refinements can be performed, which opens up a
further level of validation. This has moved from a culture of ‘having to take
Leading Edge Chemical Crystallography Service Provision and Its Impact on. . .
101
limited opportunity to grow better crystals, whereas the macromolecular crystallographer will robotically attempt to grow crystals of the same sample in highthroughput modes under as many different conditions as possible.
These factors naturally lead to a much wider range, in terms of the quality, of
crystal structures produced by modern facilities. However, the ability to validate
results to a greater extent has also progressed. Due to the increased standardisation
and digitisation of data and results, there are now not only a larger number of
quantifiable metrics to assess quality, but also they can be combined and compared
to further probe and measure quality. A couple of decades ago, a structure would be
judged by just a handful of metrics, and often the non-expert would only consider the
R-factor. However, modern validation tools are now accessible to all and results
presented in such a way that it is relatively easy to judge quality based on a whole
range of factors. So, one might now consider that it would be possible to publish a
much wider range of quality of results, if not all that can be deemed as ‘correct’. This
is a crucial change in the mindset of publishing, but a key one to support modern
chemistry requirements.
This gives rise to the concept that ‘data need to be fit for purpose’. The primary
purpose of a service crystallography structure in the past was essentially a proof of
what has been synthesised. Increasingly, over the years, crystallography has moved
on from this configuration/conformation requirement to look in detail at the nature of
bonding and crystal structures in the form of intermolecular interactions and thus
requiring a certain quality of result. So, it should be perfectly acceptable, for
example, to publish a structure that critically reveals the stereochemistry of a
compound but does not reliably provide any detailed insight into bonding. Of course,
the structure must be correct, but it can be of low quality – and these factors can
readily be assessed.
Furthermore, in the modern era, it is now critically important to consider that a
result has a very (in some cases, more?) important role when incorporated in and
contributing to the whole body of knowledge, e.g. a structure becoming a component
of a database. There are illustrations throughout the literature, and this article, of the
benefits of being able to analyse a database to derive rules and provide the basis for
informatics. In an ideal world therefore, every structure determined should in
principle contribute to this body of knowledge.
We can make a big step towards this goal. In the modern world of data science, it
is very common to spend a significant amount of time ‘cleaning’ and ‘processing’ a
large dataset. If we are to move towards a ‘data-fit-for-purpose’ approach, then
validation becomes an imperative process. This can readily be achieved not only
using our modern validation tools but also by making all of our data available. In
recent years common practice has shifted, over a relatively short period of time, from
provision of structure factors being very rare to this being close to mandatory for
most publishing routes. Deposition of structure factors has empowered a whole new
level of validation, where not only the model can be compared to the data from
which it was derived but also refinements can be performed, which opens up a
further level of validation. This has moved from a culture of ‘having to take
Leading Edge Chemical Crystallography Service Provision and Its Impact on. . .
101