Robust error calibration for serial crystallography

In macromolecular crystallography (MX), integrated intensities’ final uncertainties are typically calculated by an empirical transformation of initial counting statistics uncertainties, a procedure generally known as error calibration. Existing MX error calibration algorithms were designed for single-crystal rotational diffraction experiments. This work describes their further adaptation to serial femtosecond crystallography and is motivated by fundamental differences in data and experimentation between these methods. We propose a new parameterized error transformation that applies different levels of error to each lattice based on its correlation to a scaling reference. A maximum likelihood approach is used to robustly optimize the transformation’s parameters. This algorithm is implemented within cctbx.xfel.merge and is demonstrated to improve the accuracy of the uncertainty estimates.In macromolecular crystallography (MX), integrated intensities’ final uncertainties are typically calculated by an empirical transformation of initial counting statistics uncertainties, a procedure generally known as error calibration. Existing MX error calibration algorithms were designed for single-crystal rotational diffraction experiments. This work describes their further adaptation to serial femtosecond crystallography and is motivated by fundamental differences in data and experimentation between these methods. We propose a new parameterized error transformation that applies different levels of error to each lattice based on its correlation to a scaling reference. A maximum likelihood approach is used to robustly optimize the transformation’s parameters. This algorithm is implemented within cctbx.xfel.merge and is demonstrated to improve the accuracy of the uncertainty estimates.