On the variability of experimental data in macromolecular crystallography.

Experimental errors as determined by data-processing algorithms in macromolecular crystallography are compared with the direct error estimates obtained by a multiple crystal data-collection protocol. It is found that several-fold error inflation is necessary to account for crystal-to-crystal variation. It is shown that similar error inflation is observed for data collected from multiple sections of the same crystal, indicating non-uniform crystal growth as one of the likely sources of additional data variation. Other potential sources of error inflation include differential X-ray absorption for different reflections and variation of unit-cell parameters. The underestimation of the experimental errors is more severe in lower resolution shells and for reflections characterized by a higher signal-to-noise ratio. These observations partially account for the gap between the expected and the observed R values in macromolecular crystallography.