Weichenberger Christian X, Rupp Bernhard
Center for Biomedicine, European Academy of Bozen/Bolzano (EURAC), Viale Druso 1, I-39100 Bozen/Bolzano, Italy.
Department of Forensic Crystallography, k.-k. Hofkristallamt, 991 Audrey Place, Vista, CA 92084, USA.
Acta Crystallogr D Biol Crystallogr. 2014 Jun;70(Pt 6):1579-88. doi: 10.1107/S1399004714005550. Epub 2014 May 24.
The probabilistic estimate of the solvent content (Matthews probability) was first introduced in 2003. Given that the Matthews probability is based on prior information, revisiting the empirical foundation of this widely used solvent-content estimate is appropriate. The parameter set for the original Matthews probability distribution function employed in MATTPROB has been updated after ten years of rapid PDB growth. A new nonparametric kernel density estimator has been implemented to calculate the Matthews probabilities directly from empirical solvent-content data, thus avoiding the need to revise the multiple parameters of the original binned empirical fit function. The influence and dependency of other possible parameters determining the solvent content of protein crystals have been examined. Detailed analysis showed that resolution is the primary and dominating model parameter correlated with solvent content. Modifications of protein specific density for low molecular weight have no practical effect, and there is no correlation with oligomerization state. A weak, and in practice irrelevant, dependency on symmetry and molecular weight is present, but cannot be satisfactorily explained by simple linear or categorical models. The Bayesian argument that the observed resolution represents only a lower limit for the true diffraction potential of the crystal is maintained. The new kernel density estimator is implemented as the primary option in the MATTPROB web application at http://www.ruppweb.org/mattprob/.
溶剂含量的概率估计（马修斯概率）于2003年首次引入。鉴于马修斯概率基于先验信息，重新审视这种广泛使用的溶剂含量估计的经验基础是合适的。在蛋白质数据银行（PDB）快速增长十年后，用于MATTPROB中原始马修斯概率分布函数的参数集已更新。一种新的非参数核密度估计器已被实现，可直接根据经验溶剂含量数据计算马修斯概率，从而避免了修改原始分箱经验拟合函数的多个参数的需要。已研究了其他可能决定蛋白质晶体溶剂含量的参数的影响和相关性。详细分析表明，分辨率是与溶剂含量相关的主要且占主导地位的模型参数。对低分子量蛋白质比密度的修改没有实际影响，且与寡聚化状态无关。存在对对称性和分子量的微弱且在实际中无关紧要的相关性，但无法通过简单的线性或分类模型得到令人满意的解释。关于观察到的分辨率仅代表晶体真实衍射潜力下限的贝叶斯观点仍然成立。新的核密度估计器在http://www.ruppweb.org/mattprob/的MATTPROB网络应用程序中作为主要选项实现。
