Introduction

There are a number of potential difficulties with the refinement of macromolecular structures including the unfavorable ration of observations to parameters, the magnitude of the computational requirements, and deficiencies in the starting model ranging from small errors in the coordinates to gross errors arising from misinterpretation of the electron density map. These difficulties have led to the development of different refinement strategies, each of which has its own advantages and disadvantages (e.g. Diamond, 1971; Watenpaugh, Sieker, Herriott & Jensen, 1973; Freer, Alden, Carter & Kraut, 1975; Sussman, Holbrook, Church & Kim, 1977; Jack & Levitt, 1978; Konnert & Hendrickson, 1980; Agarwal, 1978; Jones & Liljas, 1984). In order to increase the number of observations it is usual to include knowledge of the stereochemistry of the protein. Bond distances, bond angles, planarity and limits on the approach distances of non-bonded atoms can all be specified. It can also be profitable to incorporate additional information, as in the co-refinement of bovine pancreatic trypsin inhibitor with X-ray and neutron data (Wlodawer & Hendrickson, 1982). If other data are available, such as independent phase information from isomorphous replacement, anomalous scattering, or phase information from molecular replacement, it might be desirable to include this information as well.

The package of programs described uses the principle of restrained least-squares refinement. The package is designed to be as general purpose as possible. Stereochemistry, for example is defined in a general way that can be applied to proteins, nucleic acids, prosthetic groups, solvent atoms and so on. The package has been made as efficient as possible by using the fast Fourier transform algorithm to carry out all the crystallographic transformations.

One limitation of many refinement programs is their inflexibility; portions cannot be replaced nor new functions added without extensive modification of the existing code. This limits the ability of the user to experiment with different refinement strategies. In order to modify the refinement program one must understand the data structure and algorithms of the entire program. The refinement package described here was designed to avoid this limitation. The process of refinement is broken down into basic units and an independent computer program handles each task. Each functional unit communicates with the other programs in the package by way of files of well defined format. To modify or replace any program only requires that the user understand the function of that program; the rest of the programs will function as before. In this manner calculations which can be optimized by space-group-specific algorithms (such as fast Fourier transforms) can be calculated differently for different crystal structures by a simple substitution of the appropriate program.