------------ CCP4 Newsletter - June 1996 ------------
MRC Laboratory of Molecular Biology,
Cambridge
CB2 2QH, U.K.
E-mail: andrew@mrc-lmb.cam.ac.uk
The advantages of collecting a dataset with a very high degree of completeness are well recognised. Even the loss of a few percent of the unique data can have a highly detrimental effect on molecular replacement solutions and density modification procedures. Nevertheless, it is not unusual to see a completeness of below 90%, even for the native data, in published structures. In order to simplify the task of collecting a complete dataset a new semi-automatic data collection strategy has been incorporated into the latest version of MOSFLM (5.30 or later).
The procedure assumes that data are being collected on a detector with a single rotation axis such as the Mar, R-Axis or Mac Science detectors, and is based on the OSCGEN/UNIQUE/COMPLETE set of programs originally written 12 years ago which formed part of the CCP4 suite of crystallographic software (although OSCGEN/COMPLETE have been removed from release 3.0 of CCP4). This problem has also been addressed in a general way by Nikonov & Chirgadze (1985) and a very similar approach has been described for the FAST area detector (Vickovic et al., 1994)
The program requires as input all of the parameters that are associated with processing a set of images, including the crystal symmetry and orientation which can be determined by auto-indexing a preliminary image. Given this information, the program will determine the starting phi value and the total phi rotation required to collect a dataset with the highest possible completeness. For crystal symmetries higher than triclinic, data will be lost in the cusp region unless the principle symmetry axis is inclined to the rotation axis and it is therefore an advantage to deliberately "mis-set" the crystal to eliminate the cusp region (the program will issue a warning message if this conditions has not been met).
While the Laue group symmetry determines the total rotation angle required to collect 100% complete data, in practice it is possible to collect a very high degree of completeness (>95%) with a total rotation considerably less than that required for 100% completeness (particularly if the crystal is oriented so that the rotation is not around a principle axis). For example, in an orthorhombic system it is possible to collect 95% complete data with a rotation of 60% or 90% with a rotation of 45° (in both cases using two non-contiguous segments). This can be particularly useful when collecting data at a synchrotron source and time constraints do not allow collection of the full rotation required by the Laue group. Indeed, there is a good argument for collecting data in this way even if sufficient time is available, because it minimises the risk of ending up with a seriously incomplete dataset in the event of equipment failure. Once the two (or more) most important segments have been collected, remaining beamtime can be used to fill in the remainder of the rotation range. To use this approach the user should specify the total rotation angle (eg 60 degrees) and the number of segments (up to 3) to be used. The program will then automatically determine the phi values of the segments which will give the maximum possible completeness. Normally the segments will have equal sizes (in phi) but these can be specified by the user. Alternatively the user can specify the phi segments to be generated explicitly. In either case the program will give a detailed breakdown of the completeness, and multiplicity, both as a function of resolution and rotation angle, and the percentage of Friedel pairs is also tabulated. Finally, if incomplete data has been collected for one (or more) crystal(s) and data collection is about to start on a second (or subsequent) crystal, the program can be used to determine the optimum phi values to be used for the current data collection to give the most complete final dataset.
The statistics on completeness take no account of the oscillation angle per image used in data collection and in order to achieve this completeness it is crucial to use an oscillation angle that avoids reflections being lost due to spatial overlap. An option is therefore provided which will determine the maximum possible oscillation angle that can be used while allowing a specified percentage of spatially overlapped reflections (which can be zero). (This is an improved version of the TESTGEN option in the OSCGEN program). In a future release it is planned to allow the effect of a specified data collection scheme (phi range and rotation angles per image) to be tested, but this is not possible at present.
The cpu requirements are determined by the unit cell dimensions and the resolution of the data. On an SGI Indy ( 100MHz R4600 processor) the cpu time is just over 4 minutes to determine the best two 30° segments for an orthorhombic cell 100 x 100 x 100Å and a maximum resolution of 2.5Å. For large cells or high resolution the calculation can become very lengthy. However, as the completeness depends on the volume of reciprocal space swept out during the crystal rotation, the calculation can be speeded up substantially by reducing the size of the unit cell. For example, if all the cell dimensions are halved, the time required will be reduced by a factor of almost 8. Inevitably this will introduce small sampling errors but providing the smallest dimension of the "reduced" cell is 15-20 times the maximum resolution the sampling errors will be very small. The facility to speed up the calculation has been included in the latest version of MOSFLM (5.40). The option to maximise the completeness of the anomalous data (i.e. the maximum number of anomalous pairs) rather than the unique data has also been implemented in this version.
MOSFLM is available by anonymous ftp as follows: