Brünger A T, Clore G M, Gronenborn A M, Saffrich R, Nilges M
Howard Hughes Medical Institute, Yale University, New Haven, CT 06511.
Science. 1993 Jul 16;261(5119):328-31. doi: 10.1126/science.8332897.
Structure determination of macromolecules in solution by nuclear magnetic resonance (NMR) spectroscopy involves the fitting of atomic models to the observed nuclear Overhauser effect (NOE) data. Complete cross-validation has been used to define reliable and unbiased criteria for the quality of solution NMR structures. The method is based on the partitioning of NOE data into test sets and the cross-validation of statistical quantities for each of the test sets. A high correlation between cross-validated measures of fit, such as distance bound violations and NMR R values, and the quality of solution NMR structures was observed. Less complete data resulted in poorer satisfaction of the cross-validated measures of fit. Optimization of cross-validated measures of fit will likely produce solution NMR structures with maximal information content.
通过核磁共振（NMR）光谱法测定溶液中大分子的结构涉及将原子模型与观察到的核Overhauser效应（NOE）数据进行拟合。完全交叉验证已被用于定义溶液NMR结构质量的可靠且无偏差的标准。该方法基于将NOE数据划分为测试集，并对每个测试集的统计量进行交叉验证。观察到交叉验证的拟合度量（如距离约束违反和NMR R值）与溶液NMR结构质量之间存在高度相关性。数据完整性较低导致交叉验证的拟合度量的满意度较差。优化交叉验证的拟合度量可能会产生具有最大信息含量的溶液NMR结构。
