Sophisticated and Robust Algorithms for Modelling Statistical Results of Data Collection

A particularly unique part of the work done on this topic has been the extensive use of the ground-breaking advances that took place towards the design of strategies on the basis of a statistical model of radiation damage in proteins (Partner 1A). The software BEST, developed by Partner 1A (EMBL-HH), now takes into account anisotropy in diffraction, mosaicity, disorder and, perhaps most importantly, radiation damage, for optimising instrumental parameters. This task succeeded well beyond expectation in establishing a solid statistical basis for planning experiments, which can deliver a constant I/σ(I) ratio throughout the resulting dataset. This advance was implemented in the version 3.2 of BEST. The correctness and robustness of these new developments towards the quantitative prediction of radiation damage has been extensively tested and validated by collection of real data on the ESRF ID14-4 beamlines ID14-2,4, ID23-1, ID29 and BM14 during 2007, followed by comparison of the predicted statistics with the observed ones. For a broad variety of samples, including e.g. 30S ribosomal crystals, good agreement between predicted and observed data statistics is achieved, provided the condition of sample bathing in the beam is satisfied, Figure 3.6.a. Another new option of optimising the crystal orientation for a single omega sweep has been introduced in BEST. The tests demonstrated strong dependency of the results of initial crystal characterisation on the crystal orientation, therefore the strategy procedure requires sample re-assessment in a chosen orientation. Overall, BEST implements the estimation of the attainable resolution limit for a given sample under optimal data collection parameters. The limit is given by the either the total exposure time allowed by the user or objectively by the radiation damage model, Figure 3.6.b. This measure provides a generalised estimate of the sample quality (combining the diffraction strength, mosaicity, background scattering etc.) and makes it independent on the experimental conditions selected for characterising the crystal.

Figure 3.6.a. Relationship between the radiation damage parameters (non-isomorphism and the intensity decay) forming the ground for the BEST data collection strategy prediction

Figure 3.6.b. Estimation of the optimum exposure time with no account for crystal radiation damage (black) and with proper statistically-based account (red). The latter indicates that there is the single optimum setup that will provide the best quality of the diffraction data.