Parkhurst James M, Winter Graeme, Waterman David G, Fuentes-Montero Luis, Gildea Richard J, Murshudov Garib N, Evans Gwyndaf
Diamond Light Source Ltd, Harwell Science and Innovation Campus, Didcot OX11 0DE, UK; Laboratory of Molecular Biology, Francis Crick Avenue, Cambridge CB2 0QH, UK.
Diamond Light Source Ltd, Harwell Science and Innovation Campus, Didcot OX11 0DE, UK.
J Appl Crystallogr. 2016 Oct 21;49(Pt 6):1912-1921. doi: 10.1107/S1600576716013595. eCollection 2016 Dec 1.
A method for estimating the background under each reflection during integration that is robust in the presence of pixel outliers is presented. The method uses a generalized linear model approach that is more appropriate for use with Poisson distributed data than traditional approaches to pixel outlier handling in integration programs. The algorithm is most applicable to data with a very low background level where assumptions of a normal distribution are no longer valid as an approximation to the Poisson distribution. It is shown that traditional methods can result in the systematic underestimation of background values. This then results in the reflection intensities being overestimated and gives rise to a change in the overall distribution of reflection intensities in a dataset such that too few weak reflections appear to be recorded. Statistical tests performed during data reduction may mistakenly attribute this to merohedral twinning in the crystal. Application of the robust generalized linear model algorithm is shown to correct for this bias.
提出了一种在积分过程中估计每个反射背景的方法，该方法在存在像素异常值的情况下具有鲁棒性。该方法使用广义线性模型方法，与积分程序中处理像素异常值的传统方法相比，该方法更适合用于泊松分布数据。该算法最适用于背景水平非常低的数据，在这种情况下，正态分布假设作为泊松分布的近似不再有效。结果表明，传统方法可能导致背景值的系统性低估。这进而导致反射强度被高估，并导致数据集中反射强度的整体分布发生变化，使得记录的弱反射过少。在数据处理过程中进行的统计测试可能会错误地将此归因于晶体中的merohedral孪晶。结果表明，应用鲁棒广义线性模型算法可以纠正这种偏差。
