Abstract
Macromolecular refinement is an optimization process that aims to produce the most likely macromolecular structural model in the light of experimental data. As such, macromolecular refinement is one of the most complex optimization problems in wide use. Macromolecular refinement programs have to deal with the complex relationship between the parameters of the atomic model and the experimental data, as well as a large number of types of prior knowledge about chemical structure. This paper draws attention to areas of unfinished business in the field of macromolecular refinement. In it, we describe ten refinement topics that we think deserve attention and discuss directions leading to macromolecular refinement software that would make the best use of modern computer resources to meet the needs of structural biologists of the twenty-first century.
Highlights
Macromolecular refinement is the crucial stage in macromolecular structure determination at which the structural model is improved and finalized
Every stage of the structure-determination process should take place as part of the one and only Holy Grail of software development for structural biology: a monster suite of programs that is able to carry out all stages of the structure-determination process in one
A blueprint for such a refinement program is described in Cherfils & Navaza (2019), where the authors propose the direct use of cryo-EM micrographs as the data for refinement and validation
Summary
Macromolecular refinement is the crucial stage in macromolecular structure determination at which the structural model is improved and finalized. No program implements refinement against X-ray diffraction images or cryo-EM micrographs directly (see Cherfils & Navaza, 2019): in both the fields of X-ray crystallography and cryo-EM, parameters in the data reduction and structural models have successfully been optimized in separate programs because they share very small off-diagonal terms with atomic model parameters (offdiagonal elements which already are nearly equal to zero can be ignored without a significant degradation of the speed of convergence) This is true, only for ‘complete’ data sets with large numbers of images. Even if both classes of parameters are co-varied, setting these off-diagonal elements to zero indicates to the optimizer that no connection should be expected While this problem can be cured by calculating a second-derivatives matrix block for each atom, including the correction for the symmetry-related atomic overlap, most programs only provide tools equivalent to cold medicine: ignore the cause but hide the symptoms. Nearly 50 years after the initial crystal structure was determined, it is clear that the refinement is still far from complete
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have