view article

Figure 3
Evaluating the quality of experimental projection maps by randomly varying the observed peak amplitudes. The Fc columns of the HKL files used to generate the projection maps in Figs. 1(b)[link] and 2(b)[link] were extracted using Matlab and either randomly rearranged (relative to their associated reflections and phases) or randomly kicked (modulated) by factors ranging from 50% to 300%. Ten iterations of each randomization method were performed and three of the 20 resulting maps are shown for streptavidin (a) and bacteriorhodopsin (b). File names are displayed for each iteration and `rand' represents the random rearrangement tests while `kick' represents the random modulation tests. The projection maps from the experimentally measured amplitudes `exp' are shown (bottom right) for comparison. For both streptavidin and bacteriorhodopsin, the experimentally measured projection map had the highest match (cross-correlation) to the known structure. In some of the randomized maps the correlation coefficient approaches the level seen for the experimentally measured maps (`high') but the majority of maps show significant differences in the projected density and had low correlation (`low'). Tables comparing the HKL file details corresponding to each map are shown on the right. For quick visualization of the amplitude hierarchy, the highest and second highest Fc value for each map is highlighted in the table in light red and light blue, respectively. While randomized amplitude tests that gave rise to high map correlations maintained the overall hierarchy of the experimentally measured amplitudes for low-order reflections (which strongly influence the overall density distribution), disrupting the amplitude hierarchy as seen in the low correlation maps (either through amplitude rearrangement or modulation) adversely affects the resulting projection maps. This suggests that even though a single-shot measurement with LCLS will yield structure factors with large error bars due to the fluctuating source parameters that ultimately need to be averaged over multiple crystals to converge to reliable values, the presented results clearly yield better structures than randomized Bragg intensities and therefore the intensities measured are not random.

Volume 1| Part 2| February 2014| Pages 95-100
ISSN: 2052-2525