Section 18.3.4. Future perspectives

It seems obvious to seek the best (most accurate) possible parameterization and establish it as a standard (to enable statistical structure comparisons). This does not seem to be a realistic goal for several reasons. Firstly, the parameterization is less a determinant of accuracy than the quality of the data and the method of refinement. Secondly, the quality of existing parameterization and the potential for new environment-dependent parameters improves as more structures are solved and databases grow. Such new parameters can be derived from conformation-dependent statistics (cis- and trans-proline is an example described above), hydrogen-bonding geometries etc. Finally, protein structures are generally solved not to build a statistically optimized protein database, but to discover biophysical functional mechanisms.

The growth of structural databases will improve our understanding of structural properties (Wilson et al., 1998); the highest-resolution protein structures will contribute most to the database, while low-resolution structures will profit most from improved predictive power. Structures that require restrained refinement both draw on the database for refinement parameters and integrity checks, and also contribute to it; a kind of boot-strapping procedure to re-refine deposited structures with iteratively improved parameters is conceivable (if convergent). The consequent removal of parameterization `signatures' in, e.g., bond and angle parameters seems unlikely to have practical consequences beyond identification of, e.g., catalytically relevant outliers, but qualitative improvements in structure comparison might be revealing in unexpected ways. Such an effort will require adequate computational resources and the deposition of structure factors or, even better, diffraction images.