CEITEC-Central European Institute of Technology, and National Centre for Biomolecular Research, Faculty of Science, Masaryk University Brno , Kamenice 5, 625 00, Brno-Bohunice, Czech Republic.
J Chem Inf Model. 2013 Oct 28;53(10):2548-58. doi: 10.1021/ci400448n. Epub 2013 Sep 13.
We focused on the parametrization and evaluation of empirical models for fast and accurate calculation of conformationally dependent atomic charges in proteins. The models were based on the electronegativity equalization method (EEM), and the parametrization procedure was tailored to proteins. We used large protein fragments as reference structures and fitted the EEM model parameters using atomic charges computed by three population analyses (Mulliken, Natural, iterative Hirshfeld), at the Hartree-Fock level with two basis sets (6-31G*, 6-31G**) and in two environments (gas phase, implicit solvation). We parametrized and successfully validated 24 EEM models. When tested on insulin and ubiquitin, all models reproduced quantum mechanics level charges well and were consistent with respect to population analysis and basis set. Specifically, the models showed on average a correlation of 0.961, RMSD 0.097 e, and average absolute error per atom 0.072 e. The EEM models can be used with the freely available EEM implementation EEM_SOLVER.
我们专注于参数化和评估经验模型，以便快速准确地计算蛋白质中构象依赖的原子电荷。这些模型基于电负性均衡方法（EEM），参数化过程针对蛋白质进行了定制。我们使用大型蛋白质片段作为参考结构，并使用三种荷电分析（Mulliken、Natural、迭代 Hirshfeld）计算的原子电荷拟合 EEM 模型参数，在 Hartree-Fock 水平上使用两个基组（6-31G*、6-31G**）和两种环境（气相、隐式溶剂化）。我们参数化并成功验证了 24 个 EEM 模型。当在胰岛素和泛素上进行测试时，所有模型都很好地再现了量子力学水平的电荷，并且在荷电分析和基组方面是一致的。具体来说，这些模型的平均相关系数为 0.961，均方根偏差为 0.097 e，每个原子的平均绝对误差为 0.072 e。EEM 模型可以与免费提供的 EEM 实现 EEM_SOLVER 一起使用。
